DESIGNING SERVICE FOR HUB-AND-SPOKE NETWORK Readings: J. Braklow, W. Graham, S. Hassler, K. Peck and W. Powell. Interactive Optimization Improves Service and Performance for Yellow Freight System. Interfaces 22:1, pp. 147-172, 1992 Topics: Hub-and-spoke systems Operational issues and challenges for LTL carriers Decisions, constraints, and objective function Solution strategies I. HUB-AND-SPOKE SYSTEMS What is a hub-and-spoke system? Hub-and-spoke systems facilitate the concept of economy of scale and provide a more reliable frequency of services through consolidation of shipments. A terminal is called a break-bulk (or break, hub) if it is a large terminal where consolidation takes place. The consolidation process includes unloading shipments from trucks, sorting the shipments according to their destination, and reloading the shipments to the appropriate trucks. A terminal that is not a break-bulk is called end-of-line terminal or satellite terminals. Typically, shipments in a hub-and-spoke system are going from an end-of-line terminal to another end-of-line terminal via one or two break-bulks. This is similar to the case that if you go from the airport of a small city to the airport of another small city, you need to change flights once or twice in the airports of major cities. Less-than-truckload network is a hub-and-spoke system for freight using trucks. Typical shipment size is between a few hundred pounds to over a thousand pounds where each shipment is not large enough to fill a whole container (or trailer) but is too big for express carriers. 1
Major types of long-distance road freight transportation: (A) Truckload (TL): shipments go directly from origins to destinations Destination origin (B) Less-than-Truckload (LTL): shipments go through hubs (break-bulks) 2
II. OPERATIONAL CHARACTERICS OF HUB-AND-SPOKE SYSTEMS FOR LTL The major terminals with consolidation functions are called hubs or breaks while the other terminals are called End-of-Line (EOL) or Satellites. Case of Yellow Freight System, Inc: 3
In an LTL network, we need to manage the flow of shipments (how to get from the origin to the destination), the flow of drivers (taking which segment of the transportation), and the flow of trailers. 4
There are some patterns that drivers should follow. These patterns are generally determined jointly by the company and the labor union to ensure that drivers are not over-worked and can get relatively stable income. Some common patterns are listed below. 5
SYSNET is a large-scale interactive optimization system developed at Yellow Freight System to optimize the routing of shipments and design of the network. It had produced substantial cost savings and improved the overall planning responsiveness. 6
III. DECISION MODEL FOR NETWORK DESIGN 7
A problem of shipment flow with different destinations If we do not differentiate the flow on an arc according to their different destinations, the optimization model may route a shipment to the wrong location. Therefore, we need to specify the destination of the flow on an arc. On the other hand, we need to have decision variables for loaded trailers and empty trailers. The number of loaded trailers to be put on a link indicates the frequency of services to be offered on this link. Objection function: What are the cost components? 8
Costs for moving loaded LTL trailers Handling costs at breaks Handling costs at origin end-of-lines Costs for moving truckload freight Costs for repositioning empty trailers Considerations: The difficulty is that this requirement is not linear. Suppose that if a service between a pair of terminals is offered, then a minimum of service frequency, say for example 5 trailers per week, must be maintained. Suppose that we only have a volume of 2 trailers in a particular week, we still need to send 5 trailers. Therefore, the per-unit cost of the trailers is not linear. When deciding whether a service is provided, we have to consider the limitation on the available resources. 9
The total amount of flow on an arc is the sum of the flow of shipments with different destinations. This amount is usually subject to capacity constraint (e.g. total number of trailers or drivers that can be used on this service). Furthermore, if a service is not offered, we need to ensure the total flow be zero. Flow conservation constraints are fundamental to most fleet management problems. However, these constraints are often overlooked when people with no optimization or operations research training are responsible to formulate the problem. 10
IV. SOLVING THE DESIGN PROBLEM Minimize the total of G1(y, x) + G2(z) + G3(y) + G4(x TL ) + G5(x E ) where G1(y, x): Costs for moving loaded LTL trailers (These costs depend on the design variables y and frequency of service x). G2(z): Handling costs at breaks (Theses costs, such as unloading, sorting, and reloading, depend on the outbound LTL flow at breaks.) G3(y): Handling costs at origin end-of-lines (These costs depend on the number of service links out of an end-of-line terminal. The large is the number, the higher the handling cost will be) G4(x TL ): Costs for moving truckload freight G5(x E ): Costs for repositioning empty trailers Subject to (1) Integrality of network design variables y. (2) LTL flow can only move over links where services are available (3) LTL flow to a single destination follows a tree pattern (e.g. no cycle is allow) (4) Flow conservation constraints for LTL flow (5) Flow conservation constraints for TL flow (6) Flow conservation constraints for trailers (both loaded and empty) 11
Decomposition approach The original problem is broken down into the following sub -problems which are smaller in size and also exhibit special structure, allowing the use of more efficient solution methods. Network Design Problem (NDP) with given LTL flow For fixed x, z, x TL, x E, find { y } by solving Min G1(y, x) + G2(z) + G3(y) + G4(x TL ) + G5(x E ) Subject to constraints (1) Routing sub-problem (RSP) For fixed y, find {x, z } by solving Min G1(y, x) + G2(z) + Subject to constraints (2), (3) and (4) Note: This problem can be solved as a modified shortest path problem. Truckload Freight Routing Problem (TRP) For fixed x, y, find { x TL } by solving Min G4(x TL ) Subject to constraints (5) Note: This problem can be solved as a minimum cost flow problem. Empty Balancing Sub-problem (EBSP) For fixed x, x TL, find { x E } by solving Min G5(x E ) Subject to constraints (6) Note: This problem can be solved as a minimum cost flow problem. 12
Interactive optimization: Allow users to interactively `participate in the optimization process, in particular for the higher level of decisions. Decision Level of User User Control Interaction Terminal size, location, and alignment High User initiated what if Routing loaded LTL trailers Add/drop directs User initiated what if with least-cost path suggestion Computer generated suggestions; user reviews and accepts/rejects suggestions. Routing LTL freight (the load plan) Computer generated changes; user review and override Routing empties Low User review, but no control Typical menu of suggestions IMPACT OF SYSNET Qualitative benefits Forecast new flows for next 5 years to optimize load plan professionally Identify new directs with higher flow levels and lower demand on handling facilities Direct management control over network operation 13
Managers plan and respond faster Analysts are encouraged try new ideas Reduce claims Cost savings (USD 7.3 millions direct and 10 millions indirect) 1988-89:11.6% increase of freight using direct, saving $4.7 million annually Reduce shipment handling cost, $185000 annually in depreciation expense Reduce routing trailers cost, $1 million Reduce overall transportation cost, $1.42 million Improved services Evaluate field suggestions of projects quickly and accurately Reduce over 400K shipments of late deliveries 14