Product Mixture of SKU to Pod Assignment Policy in Robotic Mobile

Fulfillment System Warehouse

Nafisha Herma Hanifha

, Shuo-Yan Chou

and Iwan Vanany

Department of Industrial Management, National Taiwan University of Science and Technology, Keelung, Taipei, Taiwan

Department of Industrial Engineering, Sepuluh Nopember Institute of Technology, Surabaya, Indonesia

Keywords: RMFS, SKU to Pod Assignment Policy, Pile-On, ABC Classification, Association Rule.

Abstract: Robotic Mobile Fulfillment System Warehouse (RMFS) was purposefully created as a part-to-picker

warehouse in response to the enormously good trend of e-commerce sales. There are numerous strategies to

boost the warehouse's effectiveness. The SKU to the pod, or product assignment policy, will be the main topic

of this study. Three situations are presented in this study: SKU to pod using random, mixed classes, and mixed

classes with affinity. The second and third scenarios are designed utilizing the Weighted Support Count. The

ideal policy to improve warehouse efficiency is then determined by comparing these scenarios using a

simulation approach. By examining the quantity of pods transported under each policy, it may be determined.

The SKU to pod approach generates a larger pile-on the fewer pods there are. Therefore, the final scenario

produces the best pile-on, with an average of 6.59 pods being carried per order. In contrast, the outcomes of

the first and second situations are 7.06 and 6.90, respectively. Even if just 8% of SKUs make up the

association's rule, the figures indicate that the pile-on of the last scenario is 7% and 5% more than the other

situations. The one-way ANOVA method is used to confirm this result.

1 INTRODUCTION

The global impact of the coronavirus has altered the

nature of business. 52% of consumers, it has been

found, steer clear of both in-person shopping and

busy places. In addition, 36% postpone going

shopping in person until they receive a coronavirus

vaccination (Bhatti, et al., 2020). As a result, one of

the most common internet activities in the world is

shopping. E-commerce revenue is expected to

increase to US$6.4 trillion by 2024 from its current

level of US$4.28 trillion in 2020. (Chevalier, 2021).

The efficiency of warehouse operations must keep

up with the expansion of e-commerce sales. The

warehouse for the Robotic Mobile Fulfillment

System (RMFS) was created especially for online

shopping. This warehouse system can use robots or

RMFS, commonly referred to as AGV (Automated

Guided Vehicles), to transport the shelves, known as

pods, to the picking stations in place of human

operators (Merschformann, Lamballais, de Koster, &

Suhl, 2019). Because the parts or goods are

transported to the picking stations before the operator

gathers the goods, this kind of warehouse can also be

referred to as a part-to-picker warehouse (Murray,

2019). Order picking is typically the task that takes

the longest in the entire warehouse compared to

others (Frazelle E., 2016). As a result, improving

order picking efficiency also improves warehouse

efficiency.

There are numerous strategies to boost the

warehouse's effectiveness. There are three tiers of

decision-making issues. These are the levels of

strategy, tactics, and operations (Merschformann,

Lamballais, de Koster, & Suhl, 2019). Product

assignment is a tactical choice that influences the

effectiveness of order-picking (Li, Hua, Huang, Sheu,

& Cheng, 2020). To achieve order-picking efficiency,

it is crucial to create a good policy for product

assignment (Silva, Roodbergen, Coelho, & Darvish,

2022). Product assignment by itself has three

difficulty areas. These are the distribution of a

product over several pods, pods to zones, and

products (SKU) to pods (Mirzaei, Zaerpour, & de

Koster, How to benefit from order data: correlated

dispersed storage assignment in robotic warehouses,

2021). The first decision problem—products to pod

or product assignment policy—will be the main focus

of this study.

Hanifha, N., Chou, S. and Vanany, I.

Product Mixture of SKU to Pod Assignment Policy in Robotic Mobile Fulﬁllment System Warehouse.

DOI: 10.5220/0012103900003680

In Proceedings of the 4th International Conference on Advanced Engineering and Technology (ICATECH 2023), pages 33-41

ISBN: 978-989-758-663-7; ISSN: 2975-948X

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

Random, dedicated, and class-based storage

assignment rules are the three most used product

assignment policies (Gu, Goetschalckx, & McGinnis,

2007).

The SKU randomly assigns the pod according to

the clear and simple random assignment policy

(Mirzaei, Zaerpour, & de Koster, how to benefit from

order data: correlated dispersed storage assignment in

robotic warehouses, 2021). A dedicated assignment

policy, on the other hand, limits the use of each

storage place to a single product. This policy will

produce the shortest picking distance possible

(Muppani & Adil, 2008). Between devoted and

random assignment policies, Chan and Chan (2011)

conducted a simulation. The outcome demonstrated

that these regulations, in turn, aid in maximizing both

system throughput and storage space usage. A smaller

warehouse may be needed when using a random

policy as opposed to one that is devoted, although

proper inventory tracking may take more work (Gu,

Goetschalckx, & McGinnis, 2007). After the products

are classified into classes based on the frequency of

orders, the class-based storage assignment policy

assigns the products (Silva, Roodbergen, Coelho, &

Darvish, 2022). This approach can produce the

highest benefits with two or three courses (Yuan,

Cezik, & Graves, 2018). Significant cost reductions

and space sharing are the next two benefits (Muppani

& Adil, 2008). (Mirzaei, Zaerpour, & Koster, The

impact of integrated cluster-based storage allocation

on parts-to-picker warehouse performance, 2021).

The cluster-based storage assignment policy is an

additional product assignment policy in addition to

those three well-liked ones. In order to reduce the cost

of inventory and material handling, this policy groups

correlated goods into clusters before assigning the

products to the pods depending on the cluster (Kim

K. H., 1993).

By grouping frequently ordered products on the

same pod, RMFS warehouses can benefit (Mirzaei,

Zaerpour, & Koster, The impact of integrated cluster-

based storage allocation on parts-to-picker warehouse

performance, 2021). This policy's implementation

greatly cuts down on retrieval time and saves order-

picking labor (Frazelle & Sharp, 1989). (Mirzaei,

Zaerpour, & Koster, The impact of integrated cluster-

based storage allocation on parts-to-picker warehouse

performance, 2021).

Nearly all cluster-based policy studies

demonstrate that, given their goals, cluster-based

policies are superior to other types of policies for

storage assignment. None of them, however, are

looking for the ideal product combination that might

be used in other instances with a similar problem. The

ideal combination of product classes on pods is called

a product mixture, which reduces the amount of

delivered pods. Higher pile-on is achieved when there

are fewer pods to transfer, which may result in fewer

AGVs being required (Merschformann, Lamballais,

de Koster, & Suhl, 2019). Pile-on is when the pods

provide the majority of the units required to complete

the orders (Merschformann, Lamballais, de Koster, &

Suhl, 2019).

Order selection effectiveness is also increased

with the right product classification (Chan & Chan,

2011). According to how frequently orders are

placed, ABC classification is typically the method

used to classify products in warehouses. However, the

second assignment decision problem—pod to

zones—is typically resolved by this classification.

Additionally, they all adhere to the "one class, one

pod" principle and make no effort to determine the

ideal Product Class Mix (Products to Pod) for each

pod. As a result, the goal of this research is to

optimize pile-on by identifying the ideal product

combination.

2 OBJECTIVE

In light of the background information provided, the

following objectives of this study might be stated:

1. Choosing the optimal product mixture percentage

for the warehouse.

2. Using a simulation method, the optimum policy for

the SKU to pod decision problem is identified.

3 LITERATURE REVIEW

This chapter will show the literature review of this

research related to robotic mobile fulfilment system

(RMFS), SKU to pod assignment, ABC

classification, and association rule.

3.1 Robotic Mobile Fulfillment System

(RMFS)

Because it accommodates several SKUs and

necessitates numerous small-quantity purchases, a

robotic mobile fulfillment system (RMFS) warehouse

is an answer to the problem of increasing e-commerce

sales (Azadeh, Koster, & Roy, 2019). Robots that

carry pods on traditional warehouse shelves with

things in the pods are the RMFS company's proposed

solution (Enright & Wurman, 2011). The RMFS

warehouse has a number of advantages. Depending

ICATECH 2023 - International Conference on Advanced Engineering and Technology

on the quantity of AGVs and SKUs, RMFS's

throughput was discovered to be higher than AS/RS

in 2016. (Beuters, Cock, Hollevoet, Dobbelaere, &

Landeghem, 2016). When the inventory is divided

over several pods, the warehouse has the right number

of stations, and the pods are refilled before they run

out, the throughput increases (Tessensohn, Roy, & De

Koster, 2020).

3.2 SKU to Pod Assignment

Three common SKU-to-pod assignment policies are

dedicated, random, and class-based storage

assignments. In addition to these three well-liked

policies, cluster-based storage assignment is another

policy for product assignment. The efficiency of

warehouse operations can be increased in a number

of ways.

In order to reduce the number of groups accessed,

Kress et al. (2016) investigated implementing a

cluster-based storage assignment mechanism in a

vertical warehouse (Kress, Boysen, & Pesch, 2016).

Chuang et al. in 2012, Bindi et al. in 2014, Wang et

al. in 2019, Li et al. in 2020, and Foroughi et al. in

2020 have all demonstrated that utilizing a cluster-

based storage assignment policy results in a reduction

in journey distance. Additionally, those researchers

implemented the policy in several kinds of

warehouses. Only two of them are used in traditional

warehouses, with the remaining ones being one-

block, one-aisle, RMFS, and movable racks

warehouses (Chuang, Lee, & Lai, 2012) (Bindi,

Manzini, Pareschi, & Regattieri, 2014) (Wang,

Zhang, & Fan, 2019) (Li, Hua, Huang, Sheu, &

Cheng, 2020) (Foroughi, Boysen, Emde, &

Schneider, 2020).

By analyzing the robot's energy usage, Li et al. in

2020 sought to reduce energy consumption in

addition to lowering trip distance in the RMFS

warehouse. Along with Li et al., Mirzaei also studied

the cluster-based implementation in RMFS and

ASRS warehouses in 2021. According to the

research, this policy is the fastest at picking orders

than any other policy.

3.3 ABC Classification

Muppani & Adil and Yuan et al. conducted research

in conventional and RMFS warehouses in 2007 and

2021, respectively, to reduce trip distance for class

rack allocation in a conventional warehouse and ABC

pod assignment in an RMFS warehouse (Yuan,

Cezik, & Graves, 2018). Along with Muppani and

Adil, rack allocation was also studied by Chan &

Chan in 2011 and Ang & Lim in 2019. Their goals

differ in that unit-load warehouses aim to reduce

travel expenses whereas conventional warehouses

want to reduce journey time (Chan & Chan, 2011).

(Ang & Lim, 2019). The most recent study, Silva et

al., 2022, found that ABC zone sizing increased the

efficiency of order picking in the traditional

warehouse (Silva, Roodbergen, Coelho, & Darvish,

2022).

3.4 Association Rule

In Yang 2022, Yang contrasts two approaches to an

association rule. Jaccard Index and Weighted Support

Count are these. The Jaccard Index evaluates

similarities and differences between simple sets and

was created by the Swiss mathematician Paul Jaccard.

This technique is also used to measure the

relationship between the objects in storage

assignment study. WSC, on the other hand, integrates

the ideas of support and lift created by Ming et al. and

reflects the relationship between any pair of products

(Chiang, Lin, & Chen, 2014). The study's findings

demonstrate that the Weighted Support Count

outperforms the Jaccard Index (Yang, 2022).

4 METHODOLOGY

The next chapter will cover the research procedures.

The steps act as a guide for the study so that it can

move forward with the goals in mind.

4.1 Data Gathering

On the basis of an investigation of 55000 past orders

with the same number of total SKUs, 10000 data

orders with 5000 SKU numbers are generated. The

order data history is then examined using @RISK

software, a Microsoft Excel add-in tool. The

distribution of the generated data is guaranteed to

match that of the historical data.

4.2 Inventory Analysis

The next phase of this research is inventory analysis,

which comes after creating new data orders. The SKU

classification procedure and the computation of the

number of slots are both included in this second stage.

4.2.1 SKU Classification

In this step, the SKU is divided into three classes

based on the order frequency. 10%, 30%, and 60% of

Product Mixture of SKU to Pod Assignment Policy in Robotic Mobile Fulﬁllment System Warehouse

all SKUs are categorized according to classes A, B,

and C using the ABC rule. However, A, B, and C each

represent 60%, 25%, and 15% of the total order

frequency, respectively.

4.2.2 Number of Slots Calculation

The SKU must be divided into three classes before

determining how many slots are required for each

SKU. The number of slots required will be the same

for SKUs categorized into the same group. The

number of slots is determined using the minimal level

inventory formula in Equation (1) (Radasanu, 2016).

The goal of determining the minimal stock is to keep

track of inventories and lower operating expenses to

prevent overstock.

Minimum Level Inventory = 𝑥



(𝜎



√

𝐿.𝑍



)

(1)

Notation:

𝑥̅



= Demand Average

𝜎



= Demand Standard Deviation

𝑍



= Z score of Service Level

L = Lead Time

4.3 SKU to Pod Assignment Scenarios

Three different possibilities make up the SKU to pod

assignment. These possibilities include random,

mixed-class, and mixed-class affinity. The specifics

of each case are described below.

4.3.1 Random Assignment Scenario

Based on the data order and inventory analysis

performed in the preceding stage, the SKUs in this

scenario are assigned at random. Despite the

unpredictability, there is a rule in this scenario: each

SKU is distributed among numerous pods.

4.3.2 Mixed-Class Assignment Scenario

To complete a mixed-class assignment scenario, three

actions must be taken. First, use Equation (2) and (3)

of Independent Event Probability to determine the

product mixture percentage. The ordering between

classes are independent of one another, hence this

formula helps calculate the number of pods for each

class combination in each pod. Therefore, the chance

of each product mixture is calculated using

independent event probability. Next, use Equation (4)

to calculate the SKU in pod ratio to ascertain the

number of slots required for each class assigned to

each pod. Finally, distribute the SKUs according to

the class.

𝑃

(

𝑋𝑎𝑛𝑑𝑌𝑎𝑛𝑑𝑍

)

=𝑃

(

𝑋∩𝑌∩𝑍

)

(2)

𝑃

(

𝑋𝑎𝑛𝑑𝑌𝑛𝑜𝑡𝑍

)

=𝑃

(

𝑋∩𝑌∩𝑍′

)

(3)

%𝑋



%𝑆𝑙𝑜𝑡



%𝑆𝑙𝑜𝑡



+%𝑆𝑙𝑜𝑡



(4)

4.3.3 Mixed-Class-Affinity Assignment

Scenario

The previous scenario has led to the final scenario.

Calculating the product mixing percentage and the

SKU in pod ratio are the first two phases in this

scenario, which are the same as the first two steps of

the mixed-class assignment. The method used to

allocate the SKU to the pod differs. The SKUs are

assigned based on the affinity between the items

based on the order data history, as opposed to the

previous case when the SKU is just assigned based on

the class.

In this instance, the third step is using

Weighted Support Count to assess the link between

each SKU and the support, confidence, and lift that

can be determined using Equations (5), (6), and (7),

respectively. To assess the degree of link between

things, support is utilized. Confidence is used to show

how likely it is that a set of SKUs will be ordered

together. Lift, on the other hand, describes the kinds

of connections between the SKUs.

𝑃

(

𝐴

∪𝐵

)

𝑎

𝑁

(5)

𝑃

(

𝐵|

𝐴

)

𝑃

(

𝐴

∪𝐵

)

𝑃(

𝐴

)

(6)

𝐿𝑖𝑓𝑡



𝑃

(

𝐵|

𝐴

)

𝑃(𝐵)

𝑃(

𝐴

∪𝐵)

𝑃

(

𝐴

)

𝑃(𝐵)

(7)

Notations and details:

P = Probability

𝑎 = frequency of SKU A and B are ordered together

Lift > 1, complementary

Lift = 1, independent

Lift < 1, substitutive

4.4 Simulation

The simulation will be run using the NetLogo

program to identify which scenario has the greatest

pile in comparison to the other two situations by

counting the number of pods visited in each scenario.

ICATECH 2023 - International Conference on Advanced Engineering and Technology

4.4.1 Simulation Layout

The NetLogo warehouse arrangement is depicted in

Figure 1 below. A picking station, a storage area, and

a replenishment station make up the layout of the

RMFS. The picking station is where the picker waits

to take items out of AGV-transported pods, the

storage area is where the pods are kept, and the

replenishment station is where the empty pods are

refilled.

Figure 1: Simulation Layout.

There are various components on it in the storage

section. Which are:

1. The items are kept in the pod.

2. The picked pod is the pod that matches the products

to the given orders.

3. The AGV robot is responsible for transporting pods

to the station for picking and refilling.

4. The aisle provides room for AGV movement.

5. The pod can be positioned in an open storage spot.

4.4.2 Simulation Parameter

This simulation makes use of a number of

assumptions, which are represented as parameters in

Table 1 below.

Table 1: Simulation Parameter.

Parameter Value

Run Len

th 24 Hours

lication 10 Re

lications

Inventor

Area 1050 Locations

Inventory Capacit

935 Pods

Empty Storage 115 Locations

Pod Batch 2 x 5 Blocks

Picking Station 6 Stations

Replenishment Station 2 Replenishment

Stations

Charging Station 7 Charging Stations

Pod Capacit

100 Slots

Number of AGV 50 AGVs

AGV Speed Without

Loa

2 m/s

AGV Speed With Loa

1,5 m/s

Acceleration 1 m/s

Time for 90˚ turnin

2,5 secon

Time for 180˚ turnin

3 secon

Time for pod lifting 4 secon

AGV to pod policy Shortest Po

4.5 Statistical Test

To verify that the simulation is accurate, a statistical

test must be run. Replication adequacy testing is done

initially to make sure there are enough replications.

The second test uses Analysis of Variance to validate

substantial changes across scenarios (ANOVA).

4.5.1 Replication Adequacy Test

Equations (8) and (9) are the formulas for the

replication adequacy test (Harrel, Ghosh, & Bowden,

2012).

𝑒=



𝑡

,









𝑠

√

𝑛

(8)

𝑛



=

(𝑍







)𝑠

𝑒





(9)

𝑒 = hw (halfwidth)

𝑡 = t value from student's t distribution table

α = confidence level

𝑠 = standard deviation

𝑛 = number of replication

𝑛

′

= estimate the number of replication

𝑛𝑛



 number of replication is sufficient

4.5.2 Hypothesis Test

The next step is to do a hypothesis test using One-

Way ANOVA after verifying that the number of

replications is adequate. This kind of ANOVA takes

into account simulations with a solitary factor. The

SKU to pod assignment policy is one element in this

study. By comparing the means of three alternative

situations and demonstrating that the results are

significantly different, an ANOVA can validate the

simulation's output.

𝐻



: 𝜇



=𝜇



=𝜇



𝐻



: At least two means are different

Product Mixture of SKU to Pod Assignment Policy in Robotic Mobile Fulﬁllment System Warehouse

5 RESULT AND DISCUSSION

This chapter will show the result and discussion of

this research starting from the data gathered to the

statistical test.

5.1 Data Gathering

From the 55000 actual data order, this phase creates a

10000 data order. To make the simulation as realistic

as feasible, the new data order must correspond to the

actual data order. Both generated and real data are

classified as having a lognormal distribution after

being checked using @RISK.

5.2 Inventory Analysis

The 5000 SKUs can be divided into ABC classes

based on the overall frequency of orders. Class A has

500 SKUs, or 10% of all SKUs, which account for

60% of all order frequency. Class B accounts for 1500

SKUs, or 30% of all SKUs, and 30% of all order

frequencies. Additionally, SKUs that make up a small

portion of the total order frequency are put into class

Table 2: ABC Classification.

Class Number of SKU

A (10%) 500

B (30%) 1500

C (60%) 3000

5000

The number of slots each SKU is based on the

quantity of orders, not the frequency, unlike how

ABC classification is determined. As a result of the

variable order quantity, each class has a varying

number of slots. Equation (1) is then applied to

specify the total number of slots. Given that the

confidence level is 95%, the Z score can be found in

Appendix 1, and the slot capacity is ten units, the

calculation below demonstrates how to estimate how

many slots there should be.

The computation reveals that even though class A

has the fewest SKUs (500 SKUs), it has the most slots

(106 slots/SKU). The majority of the overall order's

SKUs are in class A. In addition, classes B and C,

each having 1500 and 3000 SKUs, respectively, call

for 17 and 5 spaces per SKU.

Table 3: Slots Required.

Class

Number of

SKU

Slots/SKU

Total Number

of Slots

A (10%) 500 106 53000

B (30%) 1500 17 25500

C (60%) 3000 5 15000

5000

93500

According to Table 3, each class needs 53000

seats for A, 25500 positions for B, and 15000 slots for

C. This means that there are 935000 spaces in all in

the warehouse. With each pod having a capacity of

100 slots, it can be calculated that 935 pods are

required in total.

5.3 SKU to Pod Assignment Scenario

A data set must be satisfied in all three cases. The

objects kept on the pods will be one column that

varies between scenarios.

5.3.1 Random Assignment Scenario

5000 SKUs, of which the first 500 are grouped into

class A, the following 1500 into class B, and the final

3000 into class C and distributed at random into 935

pods.

5.3.2 Mixed-Class Assignment Scenario

With a total of 10,000 orders, 9304 orders have SKUs

from class A, whereas 7548 orders have SKUs from

class B, and 4627 orders have SKUs from class C.

The likelihood of orders including each class can be

calculated by dividing the total orders of each class

by the total number of orders. Equation (2) can be

used to define the mixed-class order probability

following the determination of the probability orders

for each class. The product mixture percentage in the

warehouse can then be calculated using this

likelihood. The likelihood of mixed-class orders and

the number of pods needed for each mixture are

shown in Table 4.

Table 4: Product Mixture.

Product Mixture Percentage Pods

(

ABC

)

= ABC 32,5% 306

P(ABC') = AB 37,7% 352

(

ACB'

)

= AC 10,6% 100

P(BCA') = BC 2,4% 24

(

AB'C'

)

= A 12,3% 116

(

BA'C'

)

= B 2,8% 27

P(CA'B') = C 0,8% 10

Total 100,0% 935

ICATECH 2023 - International Conference on Advanced Engineering and Technology

By a margin of 37.7%, the combination of class A

and class B orders had the highest chance. From that

percentage, 352 pods are required for the AB pod. By

32.5% or more pods, the combination of all three

classes has the second-highest probability. Then, just

A pod and AC pod were required, with 116 and 100

pods, respectively. Finally, with less than 3% in each,

just B pod, BC mixed pod, and only C pod are the

three lowest.

The number of slots for each class on each pod

must be computed using Equation (4) once the

number of pods for each mixture has been

determined. It is clear that class A, which includes the

ABC, AB, and AC pods, will always account for

more than 50% of the product combination in each

pod. A third of the ABC and AB pods and two thirds

of the BC pod are dominated by Class B. Class C, on

the other hand, makes up the least amount of each

pod's product mixture, amounting to 16%, 22%, and

37% for the ABC, AC, and BC pods, respectively.

With a capacity of 100 slots per pod, it will be simple

to determine how many slots belong to each class on

each product mixture pod. Table 5 illustrates this. The

SKUs are then allocated at random using the number

of slot rules.

Table 5: Product Combination of Pod.

Class ABC AB AC BC

A 57 68 78 0

B 27 32 0 63

C 16 0 22 37

100 100 100 100

5.3.3 Mixed-Class Affinity Assignment

Scenario

The first two steps of this scenario—determining the

product combination and the slots percentage—are

the same as they are in the second scenario. The

method used to allocate the SKU to the pod differs. In

this case, association analysis with the GoogleColab

tool must be used to discover the rules. As a result,

only 400 SKUs, or 8% of all SKUs, make up the rules.

The majority of the SKUs in Classes B and C,

however, do not have any rules because they are not

ordered together frequently enough. The 400 SKUs

must therefore be assigned close together, while the

remaining SKUs are assigned at random.

5.4 Simulation

The number of pods transported for the entire order

may be determined from the simulation output. The

average number of pods transported for 10,000 orders

can therefore be determined as the performance

indicator for the simulation outcome. Higher pile-on

are achieved when the average is lower.

5.4.1 Simulation Result Analysis

Each scenario, from replications one to ten, is

contrasted in Table 6 below. It is clear that the final

scenario, which had the lowest average number of

pods moved, produced the best results. The baseline

scenario is increased by 6.4% in the mixed affinity

scenario but only by 3.49% in the mixed scenario.

Table 6: Simulation Result.

Sim

Number of Pods/Order

Scen 1 Scen 2 Scen 3

Rep 1 7,14 6,99 6,54

Rep 2 7,06 6,76 6,46

Rep 3 7,13 6,89 6,45

Rep 4 6,92 6,90 6,72

Rep 5 7,03 7,07 6,89

Rep 6 7,20 6,94 6,54

Rep 7 6,93 6,76 6,51

Rep 8 7,03 6,78 6,74

Rep 9 7,21 6,6 6,76

Re 10 7,18 6,67 6,69

Average 7,08 6,84 6,63

The outcome indicates that SKU to pod

assignment policy's ABC classification also

influences the quantity of pods transported. Because

SKUs are too widely scattered in numerous pods

without any restrictions, Scenario 1, or the baseline

scenario, with random policy, has the highest average

number of pods transported. As a result, the necessary

pods are increased relative to other SKU to pod

assignment policies. The simulation results show that

using ABC categorization to the second and third

scenario improves warehouse performance.

It would seem hard for the third scenario to

produce the best simulation outcome, especially when

compared to the second scenario, with only 8% or

equal to 400 SKUs forming the association rules.

Because the third situation just uses association rules,

Product Mixture of SKU to Pod Assignment Policy in Robotic Mobile Fulﬁllment System Warehouse

the second and third scenarios are comparable. But

after examining the order data, it was discovered that

those 400 SKUs made up the majority of the total

order—55%. Therefore, even though just a small

portion of all SKUs are related to one another, it still

has a big impact on the pile-on if those SKUs

predominate in the overall order.

According to the simulation findings and the

research above, choose the proper SKU to pod

assignment strategy can have an impact on the

efficiency of the warehouse. The greatest strategy for

maximizing pile-on is mixed class affinity policy

when compared to random and mixed class policies.

It can also be used in real e-commerce warehouses,

where a high pile-on is necessary to increase

warehouse productivity due to the volume of orders

and the variety of SKUs in each order. SKUs and

notes have relationships, and those SKUs dominate

the overall orders.

5.5 Statistical Test

The next step is to determine whether the number of

replications is adequate after receiving the simulation

results. If it is still insufficient, more replications of

the simulation are required. On the other hand, if it is

already adequate, One-Way ANOVA is used to

confirm the outcome.

5.5.1 Replication Adequacy Test

Only ten replications have been performed due to

time constraints. Therefore, the replications adequacy

test using Equation (8) and (9) must be performed to

demonstrate that the 10 replications are sufficient.

Given that the confidence level is 95%,. The number

of replications required is at least eight times greater

when compared to the mean and standard deviation

of the simulation result. As a result, the 10

replications completed are adequate for this study.

5.5.2 ANOVA

One-way factor ANOVA must be used to validate the

simulation result after confirming that the number of

replications is adequate. The result of the hypothesis

test is shown in Figure 2 below. As can be seen, the

null hypothesis is rejected and the three scenarios are

statistically different because the p-value,

0.000000182, is less than 0.05.

Figure 2: ANOVA Test Result

6 CONCLUSION

The RMFS warehouse can be made more effective in

a number of ways. The SKU to pod assignment policy

is one of the decision-related issues. The policy with

the fewest transported pods is the best to accomplish

the greatest pile-on in the warehouse, despite the fact

that different policies result in varying warehouse

performances. In this study, three policies—

Random—baseline, Mixed Class, and Mixed Class

Affinity policy—are tested as a scenario. For each

product composition, each policy has a different set

of rules.

According to the simulation results, the final

scenario produces the best pile-on, with an average of

6.63 pods being transported per order. In contrast, the

outcomes of the first and second situations are 7.08

and 6.84, respectively. Even though just 8% of SKUs

conform to the association's criterion, the figures

indicate that the pile-on of the final scenario is 6.4%

and 3.02% larger than that of the other two situations.

However, it should be noted that 55% of the order

data is dominated by the 8% SKUs.

The one-way ANOVA is used to validate the

outcome. The three scenarios are significantly

different because their p-values are less than 0.05.

Therefore, it can be said that the SKU-to-pod

scenarios have had an impact on the effectiveness of

the RMFS warehouse. When compared to random

and mixed-class policies, the mixed-class affinity

policy is shown to be the most effective.

REFERENCES

Ang, M., & Lim, Y. F. (2019). How to optimize storage

classes in a unit-load warehouse. European Journal of

Operational Research, 278, 186-201.

Azadeh, K., Koster, R. D., & Roy, D. (2019). Robotized and

automated warehouse systems: Review and recent

developments. Transportation Science.

ICATECH 2023 - International Conference on Advanced Engineering and Technology

Beuters, K., Cock, K. D., Hollevoet, J., Dobbelaere, G., &

Landeghem, H. V. (2016). A simulation model to

compare autonomouse vehicle based warehouses with

traditional AS/RS systems. European Simulation

Conference.

Bhatti, A., Akram, H., Basit, H. M., Khan, A. U., Naqvi, S.

M., & Bilal, M. (2020). E-commerce trends during

COVID-19 Pandemic. International Journal of Future

Generation Communication and Networking, 1449-

1452.

Bindi, F., Manzini, R., Pareschi, A., & Regattieri, A.

(2014). Similarity-based storage allocation rules in an

order picking system: an application to the food service

industry. International Journal of Logistics Research

and Applications: A Leading Journal of Supply Chain

Management, 37-41.

Chan, F. T., & Chan, H. (2011). Improving the productivity

of order picking of a manual-pick and multi-level rack

distribution warehouse through the implementation of

class-based storage. Expert Systems with Applications,

2686-2700.

Chevalier, S. (2021, July 7). Global retail e-commerce sales

2014-2024. Retrieved from Statista:

https://www.statista.com/statistics/379046/worldwide-

retail-e-commerce-sales/

Chiang, D. M.-H., Lin, C.-P., & Chen, M.-C. (2014). Data

mining based storage assignment heuristics for travel

distance reduction. Expert Systems, 31, 81-90.

Chuang, Y.-F., Lee, H.-T., & Lai, Y.-C. (2012). Item-

associated cluster assignment model on storage

allocation problems. Computers & Industrial

Engineering, 63, 1171-1177.

Enright, J. J., & Wurman, P. (2011). Optimization and

coordinated autonomy in mobile fulfillmen system.

Proc. 25th AAAI Conf. Artif. Intell. (AAAI), 33-38.

Foroughi, A., Boysen, N., Emde, S., & Schneider, M.

(2020). High-density storage with mobile racks: Picker

routing and product location. Journal of the

Operational Research Society.

Frazelle, E. A., & Sharp, G. (1989). Correlated assignment

strategy can improve any order-picking operation.

Industrial Engineering, 33=37.

Gu, J., Goetschalckx, M., & McGinnis, L. (2007). Research

on warehouse operation: A comprehensive review.

European Journal of Operational Research 177, 1-21.

Kim, K. H. (1993). A joint determination of storage

locations and space requirements for correlated items in

a miniload automated storage-retrieval system.

International Journal of Production Research, 31(11).

Kress, D., Boysen, N., & Pesch, E. (2016). Which items

should be stored together? A basic partition problem to

assign storage space in group-based storage system.

IISE Transaction.

Li, X., Hua, G., Huang, A., Sheu, J.-B., & Cheng, T. (2020).

Storage assignment policy with awareness of energy

consumption in the kiva mobile fulfillment system.

Transportation Research Part E.

Merschformann, M., Lamballais, T., de Koster, M., & Suhl,

L. (2019). Decision rules for robotic mobile fulfillment

systems. Operations Research Perspectives.

Mirzaei, M., Zaerpour, N., & de Koster, R. (2021). How to

benefit from order data: correlated dispersed storage

assignment in robotic warehouses. International

Journal of Production Research

, 549-568.

Mirzaei, M., Zaerpour, N., & Koster, R. (2021). The impact

of integrated cluster-based storage allocation on parts-

to-picker warehouse performance. Transportation

Research Part E.

Muppani, V. R., & Adil, G. K. (2008). Class-based storage-

location assignment to minimise pick travel distance.

International Journal of Logistics: Research and

Applications, 247-265.

Murray, M. (2019, 11 30). Order Picking in the Warehouse.

(liveaboutdotcom) Retrieved 11 15, 2022, from

https://www.liveabout.com/order-picking-in-the-

warehouse-2221190#:~:text=Part%20to%20Picker,-

The%20part%2Dto&text=Products%20are%20moved

%20from%20the,and%20fulfills%20the%20customer

%20orders.

Silva, A., Roodbergen, K. J., Coelho, L. C., & Darvish, M.

(2022). Estimating optimal ABC zone sizes in manual

warehouses. International Journal of Production

Economics.

Tessensohn, T. L., Roy, D., & De Koster, R. B. (2020).

Inventory allocation in robotic mobile fulfillment

systems. IEEE Transactions.

Wang, M., Zhang, R.-Q., & Fan, K. (2019). Improving

order-picking operation through efficient storage

location assignment: A new approach. Computers &

Industrial Engineering.

Yuan, R., Cezik, T., & Graves, S. C. (2018). Velocity-

Based Storage Assignment in Semi-Automated Storage

Systems.

Product Mixture of SKU to Pod Assignment Policy in Robotic Mobile Fulﬁllment System Warehouse