Supply chain logistics optimization using mixed integer programming and scenario simulation

Abstract

Supply-chain routing studies can yield apparently optimal solutions that are difficult to audit when source-table compatibility, historical-cost coverage, and disruption assumptions are not reported. This study presents a reproducible decision-analytics workflow that combines a standard mixed-integer linear programming assignment model with deterministic sensitivity analysis and assumption-driven Monte Carlo stress testing. The public Supply Chain Logistics Problem Dataset was reconstructed from all seven source tables. Warehouse capacity was interpreted as documented daily order-processing capacity, and a complete H = 1–10 sweep identified the smallest cumulative-capacity multiplier that eliminated diagnostic dummy assignments. Of 9215 historical orders, 7845 (85.13%) had reconstructable freight costs; the excluded 1370 records were concentrated in a specific carrier-service-weight-band gap and were audited separately. The model generated 23,891 non-dominated real route options and solved 31,736 binary variables subject to 7864 constraints. H = 7 was the minimum zero-dummy aggregate capacity horizon, not a verified day-by-day schedule. At H = 7, optimized cost was 14,842,335.53 versus a reconstructed historical cost of 14,965,998.85, a modest reduction of 123,663.32 (0.83%). Uniform freight shocks preserved feasibility, whereas a 15% capacity loss left 248 orders unsupported. Removing dominant carrier V444_0 or origin port PORT04 left 5451 and 6610 orders unsupported, respectively. The contribution is a transparent framework for baseline coverage, aggregate-capacity feasibility, service-policy ablation, computational reporting, and route-concentration stress testing rather than a novel MILP algorithm or a general theory of resilience.