1. Introduction

Today, more than ever, the evolution of automatic storage and retrieval systems (AS/RS) is accelerating. Due to

steadily growing economic and logistic needs and the fact that more goods need to be stored, AS/RS constantly call

for further optimisations. Advanced storage systems allow automatic storage and retrieval of goods with management and control of all automated processes from one place. The two more used approaches for AS/RS

design and study in present days are analytical optimisations and simulations. Optimisation of the AS/RS operation

using modern algorithms was studied by authors such as Lerher, Šraml, Borovišek, Potrč [1] and Yang, Miao, Xue

and Qin [2]. Studies which present general overviews of warehouse design and control were presented by de Koster,

Le-Duc and Roodbergen [3], Gu, Goetschalckx and McGinnis [4] and Baker and Canessa (2009) [5]. Design of a

compact 3D AS/RS was proposed by de Koster, Le-Duc and Yugang [6], Kuo, Krishnamurthy and Malmborg,

proposed computationally efficient design conceptualization models for unit-load AS/RSs based on autonomous

vehicle technology (AVS/RS) [7] and Manzini, Gamberi and Regattieri presented a multi-parametric dynamic model

of a product-to-picker storage system with class-based storage allocation of products [8]. More specific research was

presented by Fukunari, Malmborg, Hur, Nam, Yin, Rau, Dooly, Lee and others. Fukunari and Malmborg invented

the term ‘‘interleaving’’, which refers to the pairing of storage and retrieval transactions on the same cycle to

generate DC cycle cycles [9]. Hur and Nam presented stochastic approaches for the performance estimation of a

unit-load AS/RS [10], Yin and Rau studied the dynamic selection of sequencing rules for class-based unit-load

AS/RS [11]. Dooly and Lee presented a shift-based sequencing problem for twin shuttle AS/RS [12]. In an overall

state-of-the-art review authors Roodbergen and Vis [13] found that the strength of simulation could be better

exploited in AS/RS researches by comparing numerous designs, whilst taking into account more design aspects,

especially in combination with different control policies. They proposed that sensitivity analyses on input factors

should also be performed such that a design can be obtained which can perform well within all applicable scenarios.

For our research we chose a non-traditional, unit-load, deep-lane type AS/RS. The unit-load AS/RS is typically a

large automated system designed to handle unit-loads stored on pallets or within other standard containers. The

system is computer controlled and the stacker cranes are automated and designed to handle unit-load containers. The

deep-lane AS/RS is a high density unit-load system that is appropriate when large quantities of stock are stored but

the number of separate stock types is relatively small. The loads can be stored to greater depths. Load identification

is the primary role for automatically identifying AS/RSs. The scanners are located at the inlet location, to scan a

product’s identification code. The data is sent to the AS/RS computer which, upon receipt of load identifications,

assigns and directs the load to the storage location [14]. Based on the fact that corridors (aisles) use a lot of the

available storage space and stacker cranes costs can reach up to 40% of the complete AS/RS investment, we decided

to plan an AS/RS completely without corridors and with one major stacker crane with multiple loading sites, able to

supply the complete AS/RS. Planning the layout of our AS/RS was based on a table of inquiry and the frequencies

when manufacturing individual products. Distribution of products during an AS/RS operation is dependent on factor

of inquiry (FOI), product height (PH), storage space usage (SSU) and path to dispatch (PD). Another boundary

condition included within our optimisation algorithm is that the factor of inquiry may change dynamically during

AS/RS operation regarding actual market requirements. Considering all the parameters resulted in a multi-objective

optimisation problem. We chose the particle swarm intelligence (PSO) algorithm for the optimisation as it promises

good results for complex combinatorial tasks similar to ours [15, 16, 17].

Download here