Risk-aware Balancing of Cooperative Mixed-model Assembly Lines with Correlated Task Times

This study jointly optimizes task assignments, worker crews and sequences in cooperative mixed-model assembly lines with correlated task times. A lognormal common-factor model represents within-product dependence, and a tail-bottleneck-guided adaptive neighborhood search minimizes the conditional value-at-risk of normalized cycle-time overruns under fixed staffing. Crews and orders are committed before uncertainty is realized; task starts depend on predecessor completion. Seven methods are evaluated through twenty runs on each of 48 instances extended from eight public precedence graphs. Complete enumeration and an independent SimPy model verify small-instance solutions and execution consistency. At 2000 candidate evaluations, ALNS-R reduces mean out-of-sample CVaR by 20.70% relative to deterministic balancing. Product-stream CVaR decreases by 16.08%, while normalized throughput increases by 0.56% in relative terms. Correlation and budget analyses delimit these gains, including nonsignificant comparisons and nonmonotonic scenario-count effects.