generated from DOE. (Gong et al., 2012). PSO variant used as comparison. (Packianather et al.,. 2013). DOE methodology used as a reference. (Chen and Zhu ...
RFID Tag Coverage Optimization using Particle Swarm Optimization (PSO) Algorithm and Design of Experiments (DOE) Technique Azli bin Nawawi Assoc. Prof. Dr. Khalid bin Hasnan Assoc. Prof. Dr. Sh Salleh bin Sh Ahmad Doctor of Philosophy (PhD) Faculty of Mechanical & Manufacturing Engineering Universiti Tun Hussein Onn Malaysia (UTHM)
RFID Tag Coverage Optimization using Particle Swarm Optimization
(PSO) Algorithm and Design of Experiments (DOE) Technique
Tag Coverage Optimization using Particle Swarm Optimization
(PSO) Algorithm and Design of Experiments (DOE) Technique
Radio Frequency IDentification (RFID) • Use radio wave to operate
• More efficient asset management system • Various industries: Manufacturing, retail, supply chain, construction, automotive, health care and airline.
RFID reader
RFID reader Computer
RFID tag
RFID reader Computer
RFID tag
RFID reader Computer
Detecting RFID Tags..
RFID tag
RFID reader Computer
Detecting RFID Tags..
RFID tag
RFID reader Computer
RFID tag
RFID reader Computer RFID Tag Detected Tag ID: 123xyz
RFID tag
RFID reader
RFID tag
RFID reader
RFID tag
RFID reader
20 meter
20 meter
30 meter 30 meter
30 meter 30 meter
Important Questions • How to detect all tags/as many tags as possible? • How to avoid the interference between readers? • How to get the right amount of readers? • How to make sure all readers have the equal load balance?
RFID Network Planning (RNP)
RNP Optimal Tag Coverage
Readers Interferences
Economic Efficiency
Good Load Balance
RFID Network Planning (RNP)
RNP
Optimal Tag Coverage
Readers Interferences
Economic Efficiency
Good Load Balance
RFID Tag Coverage Optimization using Particle Swarm Optimization
(PSO) Algorithm and Design of Experiments (DOE) Technique
RFID using Particle Swarm Optimization (PSO) Algorithm and Design of
Experiments (DOE) Technique
Nature Inspired Algorithms 1. Genetic Algorithm (GA). 2. Particle Swarm Optimization (PSO). 3. Bacteria Foraging Optimization(BFO).
Nature Inspired Algorithms 1. Genetic Algorithm (GA).
2. Particle Swarm Optimization (PSO). 3. Bacteria Foraging Optimization(BFO).
Nature Inspired Algorithms 1. Genetic Algorithm (GA).
2. Particle Swarm Optimization (PSO). 3. Bacteria Foraging Optimization(BFO). Simple and fast Very established Effective on various problems Fewer parameters
No. of publications related to PSO 25000
No. of publications
20000
15000
10000
5000
0 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013
Year of publication
RFID Tag Coverage Optimization using Particle Swarm Optimization
(PSO) Algorithm and Design of Experiments (DOE) Technique
RFID Tag Coverage Optimization using Particle Swarm Optimization
(
) Algorithm and Design of Experiments (DOE) Technique
PSO • No parameters settings of PSO that fits all. • More significant: PSO is used for solving complex optimization problem. • Any variation made to PSO parameters huge difference. • Parameter Tuning.
Parameter Tuning: A Weakness Area • Parameter tuning weakness area for most algorithm developers (Dobslaw, 2010). • The weak tuning of PSO parameters has the potential to generate a set of low quality results with lack of accuracy (Beielstein, Parsopoulos and Vrahatis, 2002; El-Gallad et al., 2002; Kramer, Gloger and Goebels, 2007).
• Literatures suggest the application of Design of Experiments (DOE).
RFID Tag Coverage Optimization using Particle Swarm Optimization
(PSO) Algorithm and Design of Experiments (DOE) Technique
RFID Tag Coverage Optimization using Particle Swarm Optimization
(PSO) Algorithm and (
) Technique
Design of Experiment (DOE) • A tool to understand a particular process. • All important parameters are included in the analysis. • Identify key factors quality improvement. • Identify significant parameters and their optimum settings. • Quantitative effect of each parameter.
Related Works Researcher(s)
Contribution/Outcome
(Beielstein, Parsopoulos DOE for PSO parameter tuning. Used on a and Vrahatis, 2002) very simple function (El-Gallad et al., 2002)
DOE for PSO parameter tuning. Function with one scenario only
(Tamizharasan, Barnabas and Ahamed, 2002)
Comparing 3 optimization processes. PSO and DOE are not combined
(Kramer, Gloger and Goebels, 2007)
Fractional factorial used on PSO. Only one scenario
(Chen and Zhu, 2011)
PSO variants for solving RNP. The objective function is very flexible.
(Dobslaw, 2010)
Weak parameter tuning suboptimal optimization result.
Related Works Researcher(s)
Contribution/Outcome
(Wang et al., 2011)(a)
PSO is used to optimize the mathematical model generated from DOE.
(Wang et al., 2011)(b)
PSO is used to optimize the mathematical model generated from DOE.
(Gong et al., 2012)
PSO variant used as comparison
(Packianather et al., 2013)
DOE methodology used as a reference
(Chen and Zhu, 2008)
PSO variant used as comparison
DOE for PSO’s parameter tuning is common
2 DOEs for PSO’s parameter tuning
Research Objectives Obj.1 Obj.2 Obj.3
• Construct the objective function of the RFID tag coverage optimization -> exposes the direct correlation between the parameters of RFID tag coverage and the solutions of PSO. • Justify the optimum and general settings of PSO parameters for all RNP scenarios generated by the 1st DOE session. • Evaluate the performance of the combination of PSO and DOE method against other PSO variants.
Research Scopes 1. RFID Passive tag. 2. RFID application: Asset tracking & management. 3. Original version of PSO (most stable and established). 4. Matlab (programming). 5. Minitab (DOE and statistical analysis).
Research Scopes The upper and lower ranges of each parameter of interest NO.
PARAMETER
LOWER RANGE
HIGHER RANGE
1
Number of iterations
50 iterations
200 iterations
2
Number of swarms
50 swarms
200 swarms
3
Inertia weight value
0.5
3
4
Correction factor value
0.5
3
5
Number of readers
1 reader
10 readers
6
Number of tags
10 tags
100 tags
7
Working space area
5m x 5m
30m x 30m
Research Methodology Obj.1
Construct the objective function
Obj.2
Optimum & general settings of PSO parameters
Obj.3
Evaluate the performance against other PSOs
Research Methodology Obj.1
Construct the objective function
Obj.2
Optimum & general settings of PSO parameters
Obj.3
Evaluate the performance against other PSOs
Research Methodology Obj.1
Construct the objective function
Obj.2
Optimum & general settings of PSO parameters
Obj.3
Evaluate the performance against other PSOs
Research Methodology Obj.1
Construct the objective function
Obj.2
Optimum & general settings of PSO parameters
Obj.3
Evaluate the performance against other PSOs
Research Methodology Obj.1
Construct the objective function
Obj.2
Optimum & general settings of PSO parameters
Obj.3
Evaluate the performance against other PSOs
Research Methodology Obj.1
Construct the objective function
Obj.2
Optimum & general settings of PSO parameters
Obj.3
Evaluate the performance against other PSOs
Research Methodology Obj.1
Construct the objective function
Obj.2
Optimum & general settings of PSO parameters
Obj.3
Evaluate the performance against other PSOs
Results of Objective 1
C
Coverage Optimization
(x, y) Coordinate of Reader
NT
Number of Tags
(a, b) Coordinate of Tag
Pt
Power of Reader
Ae
Effective Apperture
C
Coverage Optimization
(x, y) Coordinate of Reader
NT
Number of Tags
(a, b) Coordinate of Tag
Pt
Power of Reader
Ae
Effective Apperture
Factors used for generating RNP scenarios using DOE RNP Parameter
Low Level
High Level
Number of readers
1 reader
10 readers
Number of tags
10 tags
100 tags
Working space area
5m x 5m
30m x 30m
Sc.1
Sc.5
Sc.2
Sc.6
Sc.3
Sc.7
Sc.4
Sc.8
Research Methodology Obj.1
Construct the objective function
Obj.2
Optimum & general settings of PSO parameters
Obj.3
Evaluate the performance against other PSOs
Results of Objective 2
Sample of Results (Scenario 2) Pareto Chart of the Effects
(response is Obj. func. value, Alpha = 0.05) 59.81 F actor A B C D
B CD A BC A BD
Term
A BCD AB C AD BCD A CD D A BC AC BD
0
10
Lenth's PSE = 23.2687
20
30
40 Effect
50
60
70
80
N ame N o. of iterations N o. of sw arms Inertia w eight C orrection factor
Sample of Results (Scenario 4) Normal Plot of the Effects
(response is Obj. func. value, Alpha = 0.05) 99 D
95 90
F actor A B C D
Percent
80 70 60 50 40 30 20 10 5 1
Effect Ty pe Not Significant Significant
B
-15
-10
Lenth's PSE = 3.60277
-5
0 Effect
5
10
N ame N o. of iterations N o. of sw arms Inertia w eight C orrection factor
Sample of Results (Scenario 6) Main Effects Plot for Obj. func. value Data Means
No. of iterations
No. of swarms
1950 1900
Mean
1850 50
200
50
Inertia weight
200 Correction factor
1950 1900 1850 0.5
3.0
0.5
3.0
Optimum Settings for Each Scenario Scenario No. of Iterations
No. of Swarms
Inertia Weight Correction Value Factor Value
1
50 (min)
200 (max)
0.5 (min)
0.5 (min)
2
50 (min)
200 (max)
0.5 (min)
0.5 (min)
3
200 (max)
200 (max)
0.5 (min)
0.5 (min)
4
50 (min)
200 (max)
0.5 (min)
0.5 (min)
5
50 (min)
200 (max)
3.0 (max)
0.5 (min)
6
50 (min)
200 (max)
3.0 (max)
0.5 (min)
7
50 (min)
200 (max)
0.5 (min)
0.5 (min)
8
50 (min)
200 (max)
0.5 (min)
0.5 (min)
General Settings for All Scenarios
PSO
Number of
Number of
Inertia weight
Correction
Parameter
iterations
swarms
value
factor value
General
setting
50 (min)
200 (max)
0.5 (min)
0.5 (min)
Research Methodology Obj.1
Construct the objective function
Obj.2
Optimum & general settings of PSO parameters
Obj.3
Evaluate the performance against other PSOs
Results of Objective 3
Settings for PSO parameters for each variant of PSO PSO variant
Number of
Number of
Inertia
Correction
iterations
swarms
weight value factor value
PSO & DOE
50 (min)
200 (max)
0.5 (min)
0.5 (min)
Chen and
200
50
0.729
2.05
200
50
0.9 to 0.4
2.0
Zhu (2008) Gong, et al. (2012)
(Decreasing
value)
Sample of Results (Scenario 1) Individual Value Plot of PSO & DOE, Chen & Zhu 2008, Gong, et al. 2012 900 800 700
Data
600 500 400 300 200 100 PSO & DOE
Chen & Zhu 2008
Gong, et al. 2012
Sample of Results (Scenario 1) Boxplot of PSO & DOE, Chen & Zhu 2008, Gong, et al. 2012 900 800 700
Data
600 500 400 300 200 100 PSO & DOE
Chen & Zhu 2008
Gong, et al. 2012
Results of Objective 3 • The proposed method managed to generate lower objective function values compared to other PSO variants. • Lower value = Better result. • The fastest: PSO used in Gong, et al. (2012) due to the application of the gradually reduced the inertia weight value.
Conclusion Implementation of 2 DOE sessions on the RFID tag coverage optimization using PSO algorithm
Conclusion General setting of PSO parameters for RFID tag coverage optimization PSO
Number of
Number of
Inertia weight
Correction
Parameter
iterations
swarms
value
factor value
General
setting
50 (min)
200 (max)
0.5 (min)
0.5 (min)
Conclusion The general settings are applicable for the following scopes: NO.
PARAMETER
LOWER RANGE
HIGHER RANGE
1
Number of iterations
50 iterations
200 iterations
2
Number of swarms
50 swarms
200 swarms
3
Inertia weight value
0.5
3
4
Correction factor value
0.5
3
5
Number of readers
1 reader
10 readers
6
Number of tags
10 tags
100 tags
7
Working space area
5m x 5m
30m x 30m
Conclusion The proposed method (PSO and DOE
combination) can be considered as a robust and efficient optimization system because it manages to generate high quality results in overall RNP scenarios.
Contribution (Obj.1) Useful for beginners in the field of RFID tag
coverage optimization offers a useful reference for initiating RNP optimization.
Contribution (Obj.2) Suggests the optimum setting for PSO
parameters optimization process on every RNP scenario.
Contribution (Obj.3) Shows that the proposed method
performs better than the other PSO variants proof of the usefulness.
Research Continuity • 5 FYPs (Final Year Projects). • Research Group: – RFID Network Planning (ISE: Industrial Systems Engineering). – RFID for Warehouse Management (MES: Mechatronics and Embedded Systems).
• Industrial Case Study (Unijoh Sdn. Bhd.)
Thank you! For your precious time..