Automotive Just-In-Time (JIT) supply chains face three intertwined challenges: stochastic lead-time uncertainty, non-trivial 3D cargo loading and stability constraints, and conflicting cost-CO2-resilience objectives. Existing approaches typically address these dimensions in isolation, relying on deterministic Mixed-Integer Linear Programming (MILP) planners, purely predictive Machine Learning (ML) models, or simulation-based Digital Twins (DTs) that lack both an embedded prescriptive optimization layer and a principled data-quality safeguard. This paper presents a hybrid DT framework integrating supervised ML, autoencoder-based anomaly detection, and MILP within a discrete-time logistics simulator, instantiated on a likely real automotive JIT scenario that combines real-world accident records with representative operational parameters. An XGBoost-based module estimates hourly delay profiles from large-scale roadaccident data, while an autoencoder-based Data Quality Guard filters corrupted IoT and environmental readings before they reach the predictive pipeline. Delay profiles feed a hierarchical planner composed of Material Requirements Planning (MRP), Transportation Management System (TMS), and Warehouse Management System (WMS) layers: the TMS layer instantiates a MILP model with Big-M rescue logic for fleet allocation and emergency shipments, while the WMS layer enforces 3D bin-packing and load-stability constraints. Implemented in the SmartJIT Zero-Stop Logistics platform, the framework is validated via Monte Carlo simulation across throughput scenarios from 6 to 70 Jobs Per Hour (JPH), demonstrating robust scalability, autonomous road-to-rail modal shift, up to 70% CO2 reductions versus full-truckload benchmarks, and statistical zerostockout at a controlled cost premium.

A Hybrid Digital Twin for Resilient Just-In-Time Logistics: An Automotive Supply Chain Application

Epicoco, Nicola
2026-01-01

Abstract

Automotive Just-In-Time (JIT) supply chains face three intertwined challenges: stochastic lead-time uncertainty, non-trivial 3D cargo loading and stability constraints, and conflicting cost-CO2-resilience objectives. Existing approaches typically address these dimensions in isolation, relying on deterministic Mixed-Integer Linear Programming (MILP) planners, purely predictive Machine Learning (ML) models, or simulation-based Digital Twins (DTs) that lack both an embedded prescriptive optimization layer and a principled data-quality safeguard. This paper presents a hybrid DT framework integrating supervised ML, autoencoder-based anomaly detection, and MILP within a discrete-time logistics simulator, instantiated on a likely real automotive JIT scenario that combines real-world accident records with representative operational parameters. An XGBoost-based module estimates hourly delay profiles from large-scale roadaccident data, while an autoencoder-based Data Quality Guard filters corrupted IoT and environmental readings before they reach the predictive pipeline. Delay profiles feed a hierarchical planner composed of Material Requirements Planning (MRP), Transportation Management System (TMS), and Warehouse Management System (WMS) layers: the TMS layer instantiates a MILP model with Big-M rescue logic for fleet allocation and emergency shipments, while the WMS layer enforces 3D bin-packing and load-stability constraints. Implemented in the SmartJIT Zero-Stop Logistics platform, the framework is validated via Monte Carlo simulation across throughput scenarios from 6 to 70 Jobs Per Hour (JPH), demonstrating robust scalability, autonomous road-to-rail modal shift, up to 70% CO2 reductions versus full-truckload benchmarks, and statistical zerostockout at a controlled cost premium.
2026
digital twin, MILP, JIT logistics, supply chain resilience, sustainable transportation
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/20.500.12572/39428
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
social impact