Disassembly sequencing using genetic algorithm
Aug. 29, 2017
Presented by Yooney Cho
International Journal of Advanced Manufacturing Technology (2006) 30: 497-506
Contents
1. Introduction
2. Literature review
3. Overview of genetic algorithm
4. Numerical example
5. Conclusions
Introduction
• Reuse
• Recycle
• Remanufacturing
• Collection
• Disassembly
• Sorting
• Cleaning
• Testing
• Reworking
• Reassembly
EOL(End Of Life) processing
Literature Review
Author00 Year Solution approach
Taleb and Gupta 1997 Proposed algorithms for scheduling the disassembly of discrete and well defined product structures
Lambert and Gupta 2002 Addressed the problem of demand driven disassembly using a tree network model
Keung et al. 2001 Applied a multi-objective GA approach to a tool selection model.
Overall objective of the model was to minimize the processing time.
Loughlin and Ranjithan 1997
Proposed a GA method, so called neighborhood constraint method and concluded that the GA performed better in multi-objective problems compared to single objective problems.
Lazzerini and Marcelloni 2000
Used GA in scheduling assembly processes. They Employed a
modified partially matched crossover(PMX) and mutation operations to obtain the near optimal sequence. The precedence relationships were not considered in this model.
Bierwirth and Mattfeld 1999
Proposed a method to overcome this problem by introducing the precedence preservative crossover technique for scheduling problems.
Literature Review
Literature Review
Disassembly problem (Tree network model, goal programming…)
Quick, cost effective, reasonable solutions
Genetic Algorithm
Lazzerini and Marcelloni [24] - Not considered precedence relationships Seo et al. [26] - *Penalizing the string to eliminate infeasible strings
Regular GA is not suitable for considering precedence relationships
Bierwirth et al. [27], Bierwirth and Mattfeld [28]
– introduced *PPX to a single objective job shop scheduling problem
– This method preserves the precedence relationships during the crossover of GA
Disassembly sequencing using Genetic Algorithm (PPX method)
*PPX: Precedence Preservative Crossover
Literature Review
Overview of genetic algorithm
Problem description
Objective function
Minimizing the make-span
Decision Variables
The disassembly sequence of the components
Assumption
1) The direction of each operation is independent of the adjacent components.
2) Neither non-destructive nor destructive disassembly destroys the adjacent components.
3) Complete disassembly is required.
each component will be separated from each other.
Overview of genetic algorithm
Step1 : Initial population
Randomly created, feasible
The structure of chromosome consists of five components (sequence, direction, method, demand, material)
Step2 : Fitness evaluation
Make-span
Step3 : Selection
Half of the best-fit chromosomes are selected for the next generation
Step4 : Offspring generation
Precedence preservative crossover
Reciprocal exchange mutation
Step5 : Stop criterion
When the maximum number of generations are exceeded or no further improvement is obtained
Solution approach
Overview of genetic algorithm
Nomenclature
j: Index for component Ch: Index for chromosome gn: Index for generation pop: Index for population
Seq: Index for disassembly sequence(1-9)
n: number of components in the EOL product ncr: Number of chromosomes in the population C: conversion constant
chl: chromosome length rnd: Random number(1-9)
ctj,seq: penalty for direction change for disassembling component j in sequence seq dtj,seq: Time required to disassemble component j in sequence seq
TJ,seq: Cumulative disassembly time after component j in sequence seq is disassembled dej,seq :type of demand for component j in sequence seq
maj,seq:Material type of component j in sequence seq
mtj,seq:Penalty for disassembly method change for disassembling component j in sequence seq
F(ch,gn) : Fitness value for chromosome ch in generation gn MS(ch,gen): Total make-span of chromosome ch in generation gen MT: Total penalty for disassembly method change
CT: Total penalty for direction change
Overview of genetic algorithm
Components structure
Precedence relationships
Component 1 or 2 must be disassembled prior to any other components
Component 6 must be disassembled prior to Components 4 and 5
Component 7 must be disassembled prior to components 6 and 3
product
1 2
0 8 7 9
6 3
4 5
1
2 0 8
7 9
6 3
4 5
Overview of genetic algorithm
Example
5 sections
Sequence
Direction
Method
Demand
Material
Chromosomes(=string)
Overview of genetic algorithm
Component 1 or 2 >>any other components Component 6 >>Components 4 and 5
Component 7 >>components 6 and 3
처음부터 10개가 완벽히 feasible solution이 나오기는 어렵기 때문 에, Infeasible solutions 들은 제외 시켰을 가능성이 있음.
Initial population
Example
Repetitive Random selection by using Table 1(input data)
Feasible solutions(all precedence relationships, constraints are satisfied)
ncr(Number of chromosomes)=10
Ascending order(fitness)
Overview of genetic algorithm
Crossover
Precedence Preservative Crossover(PPX) method
Example
Randomly filled vector 1: drawn from parent1 2: drawn from parent2 Pass on the precedence relationship.
no new relationships are introduced.
Overview of genetic algorithm
Crossover
Precedence Preservative Crossover(PPX) method
Example (Child
1, detail procedure)
2 8 7 1 0 6 3 9 5 4
Chromosome 1 (Parent 1)
Chromosome 2 (Parent 2)
1 2 7 6 3 4 0 8 5 9
Mask for child 1
2 2 2 1 1 2 2 2 1 2 1 2 7 8 0 6 3 4 9 5
Child 1
Overview of genetic algorithm
Mutation
Mutate with a given probability
Precedence relationships are preserved
Not selected for mutation Selected for mutation
Preserve solutions in the current population
Exchange component 1 and 2 Randomly selected chromosomes
Overview of genetic algorithm
Fitness evaluation
☞ Fitness function (Minimize)
(=Total time of disassembly) (=Make Span)
Basic disassembly time (DT)
Penalty time Tool change time (MT) Direction change time (CT)
• Do not penalize last disassembly sequence component
• Do not penalize “Recycling pair”
*Recycling pair: two adjacent components made of same material and if they are both demanded for recycling.
☞ Fitness value = C – MS, C>MS
(Maximize)
Overview of genetic algorithm
Fitness evaluation (detail)
Ex: +x to +x is 0
Ex: N to N is 0, N to D is 1
Sequence position 0~8
n: product내의 component 수=0~9=10개
Sequence position 9
J: component
Overview of genetic algorithm
Fitness evaluation (detail)
Component j의 sequence 에서의 demand type
= 0:no demand, 1: reuse, 2: recycle
Component j의 sequence 에서의 Material type
= A:Al, S: steel, P: plastic
T
seq+1=(T
seq-1)+dt 가 옳다고 생각함.
*T : Cumulative disassembly time after component j in sequence is disassembled
Sequence-1 Sequence Sequence+1
Al steel steel
Recycle Recycle
Reuse
T
seq-1T
seq+1Overview of genetic algorithm
Fitness value calculation
Example
MS=10s+2s+25s=37s, Let C=100.
Overview of genetic algorithm
Termination
When the maximum number of generations are exceeded
Ex: 100 generations
No further improvement is obtained
Ex: If average fitness value of new generation is lower or equal to
previous generation
Overview of genetic algorithm
Genetic algorithm model steps
Initial population
Fitness calculation
Ascending order(fit)
Select first half of the population Ex: 10개
Ex: 5개
Crossover probability
PPX operation
Generate children
Ex: 5개 Clone the
selection Ex: 5개
Ex: 10개
Mutation probability Not selected for
the mutation
New generation Ex: 10개
Ex: 10개
Ex: 6개
Ex: 10개
Selected for the mutation Clone the
chromosomes
Operate Mutation
Ex: 4개
Ex: 6개 Ex: 4개
Ex: 10개
Numerical example
Input data
Numerical example
After 6 generation
Result
Numerical example
Comparison
Genetic Algorithm(GA) method
Fitness value=71
Conversion constant (c)=100
Total make-span=100-71=29 s
Code: ANSI-C
Exhaustive search (ES) method (= Brute-force search)
Fitness value=71
Conversion constant (c)=100
Total make-span=100-71=29 s
Code: ANSI-C
Obtained total 12 sequences
*ES method : Also known as Brute force search, an approach which explore the entire search space, testing every possible candidate solution