In progress Mobile Fulfillment Systems

Reference:
ERIM PhD 2012 LIS 06 RdK_YY

Abstract

Mobile fulfillment systems are a new form of automation that hold great promise for warehouses and distribution centers. As a revolutionary material handling technology that fundamentally differs from previous systems, mobile fulfillment is interesting for warehouse managers since it can greatly increase picking efficiency.
Furthermore, mobile fulfillment offers a rich area for scientific exploration in Operations Research as mutually dependent automated decision problems have to be solved in a dynamic environment. Research in mobile fulfillment will be
valuable to managers by providing strategic insights into the relationship between design configuration, operating policies, and operational and serviceperformance. Such fundamental insights can be used to build optimal designs and
policies.



Keywords

warehousing; robots; mobile fulfillment; kiva; semi-open queueing network; material handling

Time frame

2012 - 2016

Topic

Storage facilities, including warehouses, distribution centers and container terminals, can be found everywhere. They form the key nodes in supply chain networks decoupling demand from supply in time and quantity. They serve as the main points from which transport is organized, as value adding points to postpone manufacturing, to create assortments for customers, and as the main returns receiving, sorting and handling points. Traditional storage facilities occupy much space, as many products have to be stored, usually in single-deep racks separated by transport aisles. Compact systems store products multi-deep with only few transport aisles. They are usually fully automated and higher than traditional facilities. Therefore their footprint (also their environmental footprint) is smaller than single-deep warehouses. Although much research has been carried out on traditional, single-deep automated storage and retrieval systems (AS/RS), see Hausman et al. (1976), Bozer and White (1984), or the overview paper of Roodbergen and Vis (2009), research in multi-deep, or compact, storage systems is still new. We are particularly interested in unit-load systems (i.e. pallets or totes), where each load can be retrieved individually, albeit sometimes by reshuffling other loads.
The following types of compact storage systems can be distinguished:
1. Storage/Retrieval machine (crane) with a satellite linked to the crane. In this system, the crane is responsible for transporting unit loads in the x and z directions. The shuttle can drive into the multi-deep storage lanes and is responsible for the depth (y-) movement of unit loads.
2. Storage/Retrieval machine (crane) with pairwise rotating conveyors taking care of the depth movement of unit loads.
3. Very-high density storage systems, or live-cube storage systems, where each load is stored on a shuttle that can move the load in the x - or y- direction (as long as an empty space is available next to the load). The vertical movement between storage levels is carried out by a lift.
4. A lift for vertical movement in combination with one or more shuttles for the x and y- movement of unit loads. Not every load has its own shuttle. Such systems are generally called Autonomous Vehicle-based Storage and Retrieval (or AVSR) systems, as the vehicles are not linked to a crane or to the loads. Two variants exist, one where x- and y-direction transport are carried out with different vehicles and one where vehicles can move in two directions (on rails in the system).
5. Other compact storage systems, such as the Autostore (Swisslog, 2011a, 2011b) or the Kiva system (Wurman et al., 2007, Enright and Wurman, 2011).

The first group of systems are the most common systems found in practice. They are supplied by several manufacturers and system integrators, such as Westfalia and Dematic. A variant of this system is the system with a warehouse truck instead of a crane, such as those supplied by Jungheinrich and Toyota Material Handling. Instead of truck-bound satellites variants exist where a truck or S/R machine can work in combination with multiple shuttles. The truck or S/R machine is still necessary to move the shuttles to a different storage lane. Although they are most common in practice, they have not been studied in AS/RS literature (see Roodbergen and Vis, 2008). Usually, one multi-deep storage lane is used to store one product type only. Zaerpour et al. (2010) study the assignment of loads to storage locations in a cross-dock facility, such that loads of different products can share a storage lane. In the cross-dock facility the destination (i.e. departure truck and departure time windows) are known. The objective is to minimize the total makespan for retrieving all the loads.
The second group of systems is rare in practice due to the high cost, but has been studied in several papers (De Koster et al., 2008; Yu and De Koster, 2009a, 2009b, 2010, 2012). These systems are complicated in operation and control, since the crane and the multiple conveyors can work independently to a high degree. All conveyors can simultaneously presort loads to be retrieved next, or empty locations to be replenished to the crane pick up/ drop off position, while the crane is working on another load. It is possible to estimate the makespan of a group of retrieval loads (Yu and De Koster, 2012), or to determine the response time of the system under different storage strategies, like random (De Koster et al., 2008), class-based (Yu and De Koster, 2009b), or full-turnover based (Yu and De Koster, 2009a).
The third group of systems, live-cube storage systems, is very new. Applications can be found particularly in automated car parking garages in Asia (Eweco, 2011; Hyundai (2011; Automotion Parking Systems, 2011; Wohr, 2010; OTDH, 2011). They have been studied first by Gue (2006) and Gue and Kim (2007). Their operation is comparable to a Sam Lloyd´s 15 puzzle, where the idea is to move 15 tiles in a 4 by 4 square to create a picture. The objective may be to do this in a minimum number of moves (i.e. minimize response time). However, in a live-cube storage system there may be multiple empty slots per storage levels and in addition there is a shared lift connecting all storage levels, which all can work independently. The 3D version of the puzzle is therefore quite a bit more complex than the 2D version. Zaerpour et a. (2011a, 2011b) have determined the response time in such a system for random and two-class based storage of unit loads.
The fourth group of AVSR systems is booming in practice, because they are economically attractive. They combine high throughput with a very compact (hence cost efficient) construction. Several manufacturers have developed or are developing such systems (Vanderlande Industries, Savoye, Swisslog, Smoov system) with several dozens of applications in Europe. Several authors have studied such systems (Malmborg, 2002; Fukunari and Malmborg, 2009; Kuo et al., 2007; Heragu et al. 2011; Zhang et al., 2009; Roy, 2011). In these studies vehicles can move at each storage level both in the x- and y-directions. However, they move in aisles and loads are stored only single deep. Most authors use queuing network approximations to derive performance measures such as system throughput and response time, or mean and variance of throughput time of a request. These measures are used to optimize the system configuration (Roy, 2011).

In this research, the emphasis is on AVSR systems, however, with loads stored multi-deep. The vehicles (shuttles and transfer cars) can either travel in two horizontal directions or different types of vehicles are used for each of the directions. These systems are quite cost-efficient. According to the supplier of the Smoov system they have about the same cost per pallet position, as a conventional AS/RS (about EUR220; system size: about 2000-3000 pallets, and 2 or 3 shuttles; depending on configuration), but in addition, the building can be much smaller and the throughput is higher. Several questions need to be answered, such as:
o What is the response time, throughput time and variance of the throughput time of a load for a given configuration?
o What is the impact of storage strategies on performance?
o How to sequence loads in order to maximize performance?
o What is the best configuration (i.e. length, width, height) with respect to performance?
o Which system configuration is the most economical?

Supervisory Team

René de Koster
Professor of Logistics and Operations Management
  • Promotor
  • Daily Supervisor
Debjit Roy
Associate Professor of Logistics and Operations Management
  • Copromotor