Intelligent Solution System Towards Parts Logistics Optimization

Yaoting Huang, Boyu Chen, Wenlian Lu, Zhong-Xiao Jin, Ren Zheng

TL;DR

Researchers from Fudan University and SAIC Motor have developed a comprehensive optimization system for complex "parts-to-plant" logistics. By combining a specialized Tabu Search algorithm with a novel "shipment bundling" technique, the system automates the 2D loading and routing of thousands of shipments, significantly reducing total mileage compared to traditional manual planning.

What Problem Does This Paper Address?

In the automotive industry, logistics is a high-stakes jigsaw puzzle. Thousands of parts must be collected from diverse suppliers and delivered to assembly plants within strict time windows. This is a variant of the Vehicle Routing Problem (VRP), but with several real-world "headaches" added:

  1. 2D Loading Constraints: Parts aren't just points; they are bins that must be packed into a truck's floor space without overlapping and in a specific sequence (to allow forklift access).
  2. Heterogeneous Fleet: Trucks come in different sizes and models.
  3. Multi-Depot & Time Windows: Trucks start from various yards and must hit specific "docks" at suppliers/warehouses during narrow operational windows.
  4. Queuing: If too many trucks arrive at a dock simultaneously, they must queue, wasting time.

The authors call this the 2-Dimensional Loading Capacitated Multi-Depot Heterogeneous VRP with Time Windows (2L-VRPTW). Before this system, SAIC Motor relied on "old-fashioned" manual planning, which struggled to handle the combinatorial explosion of these constraints.

Methods

The authors propose a three-stage "Intelligent Solution System":

  • Initialization: To start with a feasible plan, the system imitates manual logic. It uses a community detection algorithm to cluster suppliers geographically and then assigns shipments to trucks in a descending order based on distance.
  • Optimization (The Core): They employ a Tabu Search (TS) algorithm. To prevent the search from becoming too slow due to the massive number of possible moves, they introduced a bundle technique. Instead of moving individual shipments between trucks, they group shipments from the same supplier to the same destination into "bundles." This reduces complexity and helps the algorithm escape local minima.
  • 2L-VRPTW Solver: An auxiliary solver handles the specific 2D packing and route pruning. It uses a tree-search method to find routes and a heuristic 2D packing algorithm (2PP) to ensure bins actually fit in the truck.
  • Post-optimization: This stage performs "cleanup," such as replacing large trucks with smaller, cheaper ones or merging sub-routes (solving a mini-Traveling Salesman Problem) so trucks don't return to the yard unnecessarily.

Datasets

The system was evaluated using real-world industrial data provided by SAIC Motor's logistic division:

  • 311 shipments consisting of 54 variants of bins.
  • 45 suppliers and 8 plant warehouses.
  • The production system (SPRUCE) is designed to scale up to 10,000 shipments daily across 500+ suppliers.

Key Insights and Ideas

  • Bundling is the Secret Sauce: The most significant "aha moment" is that moving individual shipments in a TS algorithm is computationally expensive and often leads to "shallow" improvements. By moving bundles, the algorithm takes larger "steps" in the solution space, leading to much faster convergence and better global optima.
  • Manual Logic as a Seed: While manual planning is inefficient for final optimization, it provides an excellent "feasible" starting point. Using community detection to mimic how humans group suppliers ensures the algorithm doesn't start from a chaotic, impossible state.
  • The Conflict of Constraints: The authors noted that optimization and post-optimization sometimes "compete" for time-window space. A route that looks shorter on paper might be impossible because it forces a truck into a dock during a peak queuing period.

Key Results

The experimental results demonstrate a massive improvement over the baseline:

  • Mileage Reduction: The initial feasible scheme (manual-like) resulted in a total mileage of 1131.80. After the Tabu Search with bundling (TS-WB) and post-optimization, the mileage dropped to 482.50—a reduction of over 57%.
  • Efficiency of Bundling: Without the bundling technique (TS-NB), the algorithm stalled at a mileage of 1014.60, proving that the bundle technique is essential for navigating the complex constraint landscape.
  • Operational Speed: The implemented system, named SPRUCE, can process 2,000 shipments in roughly 10 to 15 minutes, making it viable for daily operational planning in a fast-paced manufacturing environment.

Limitations and Future Work

While the system is highly effective, the authors identify two main areas for improvement:

  1. Computational Speed: As the number of shipments grows toward 10,000, even 15 minutes might become a bottleneck.
  2. Advanced Packing: The authors are currently exploring Deep Q-learning to solve the 2D packing problem (2PP) more efficiently than current heuristics.
  3. Parallelization: They plan to incorporate genetic algorithms that can take advantage of parallel computing to explore multiple solution paths simultaneously.

Summary updated

Abstract

Due to the complication of the presented problem, intelligent algorithms show great power to solve the parts logistics optimization problem related to the vehicle routing problem (VRP). However, most of the existing researches to VRPs are incomprehensive and failed to solve a real-work parts logistics problem. In this work, towards SAIC logistics problem, we propose a systematic solution to this 2-Dimensional Loading Capacitated Multi-Depot Heterogeneous VRP with Time Windows by integrating diverse types of intelligent algorithms, including, a heuristic algorithm to initialize feasible logistics planning schemes by imitating manual planning, the core Tabu Search algorithm for global optimization, accelerated by a novel bundle technique, heuristically algorithms for routing, packing and queuing associated, and a heuristic post-optimization process to promote the optimal solution. Based on these algorithms, the SAIC Motor has successfully established an intelligent management system to give a systematic solution for the parts logistics planning, superior than manual planning in its performance, customizability and expandability.
BibTeX
@inproceedings{huang2019intelligent,
  author = {Yaoting Huang and Boyu Chen and Wenlian Lu and Zhong-Xiao Jin and Ren Zheng},
  title = {{Intelligent Solution System Towards Parts Logistics Optimization}},
  booktitle = {WCGO},
  year = {2019},
  url = {https://openreview.net/forum?id=IVLfTpQZN2N},
}

Loading related content...

Figure