Intelligent Digital Twin for Optimizing Warehouse Operations: Embedded Optimization Components for Enhanced Order-Picking Efficiency
Abstract
The significant growth in e-commerce has highlighted the importance of optimizing warehouse operations, particularly the order-picking process, which can substantially reduce operational waste. Order-picking consists of up to 55% of total warehouse operating costs. Decision-making in this field has become more complex for human management due to varying system conditions, a large number of variables and constraints, and huge data volumes. Digital twin (DT) is a cost-effective solution that can accurately simulate system behaviors and replicate essential functional characteristics of physical system operations by utilizing real-time data. By incorporating human factors, DT plays a key role in facilitating the transition to Industry 5.0 (I5.0). As the cyber component of Cyber-Physical Systems (CPS), DT employs IT technologies, including Artificial Intelligence (AI), to simulate real-world systems and optimize complex problems. By incorporating optimization and AI-based techniques, DTs facilitate data analysis and accurate simulations, enabling more efficient decision-making. An intelligent warehouse digital twin can simulate the order-picking process, generate improvement scenarios, and assist with the implementation of effective solutions without trial and error. This study aims to develop a warehouse digital twin using a Discrete Event System Simulation model to predict and enhance operational efficiency in manual picker-to-part systems. The intelligent digital twin integrates optimization components to improve warehouse performance. The order-picking operation is modeled as a multiple-server queuing system using near real-time data. Optimization components include a dynamic order-batching (DOB) algorithm, and multi-objective storage location assignment problem (SLAP) optimizers, considering practical factors in the order-picking process. Optimization models are developed using mixed integer programming (MIP) techniques. The DOB combines heuristic and mixed integer linear programming (MILP) models, with feasibility assessed by a CP-SAT solver. Near-optimal picking tours are derived from the Travelling Salesman Problem (TSP) model, solved by a meta-heuristic optimizer. The SLAP model is developed using an integer non-linear programming (INP) technique and solved by a customized optimizer inspired by Simulated Annealing and Basin-Hopping algorithms. Experimental evaluation using warehouse data from a commercial facility in Australia shows that employing DOB reduces order throughput time by up to 15.2%, improving traveling time and distance. Multi-objective SLAP optimization enhances picking efficiency and significantly improves safety and rack stability by over 90%. These outcomes provide invaluable information for decision-makers to optimize warehouse operations without trial and error.