Determining optimum Genetic Algorithm parameters for designing manufacturing facilities in the capital goods industry

1
Determining optimum Genetic Algorithm parameters
for designing manufacturing facilities in the
capital goods industry

• Dr Christian Hicks
• University of Newcastle upon Tyne

http//www.staff.ncl.ac.uk/chris.hicks/presindex.h
tm
2
Layout literature

• Two main themes
• Facilities layout problem (FLP)
• Group Technology / Cellular Manufacturing

3
Facilities Layout Problem

• The determination of the relative locations for,
  and the allocation of available space among a
  number of workstations (Azadivar and Wang,
  2000).
• Block layouts represent resources as rectangles
• FLP formulated as quadratic set covering
  problem, mixed integer programming problem and a
  graph theoretic problem.
• The FLP involves the solution of inefficient
  NP-complete algorithms. The longest time for
  solution increases exponentially with problem
  size.
• A lot of research based upon small or theoretical
  situations.

4
Cellular Manufacturing

• Clusters of dissimilar machines are placed close
  together
• Manufacturing cells design steps
• Job assignment
• cell formation
• layout of cells within plant
• Layout of machines within cells
• Transportation system design
• 3 approaches to cell formation part family
  grouping, machine grouping and machine-part
  grouping.
• Cell formation and the layout problems are both
  NP-complete problems.

5
Cellular Manufacturing

• CM can reduce set-up and flow times, transfer
  batch sizes and WIP.
• However
• 8/9 simulation studies found that functional
  layouts performed better than CM in terms of a
  range of evaluation criteria
• 14/15 empirical studies revealed CM produced
  significant operational benefits.
• Possible explanation
• CM facilitates teamworking and provides a
  starting point for JIT. This may explain the
  difference in results obtained by research based
  upon simulation and empirical studies.

6
GA Procedure

• Use GAs to create sequences of machines.
• Apply a placement algorithm to generate layout.
• Measure total direct or rectilinear distance to
  evaluate the layout.
• Two approaches
• Algorithm can treat layouts as a single
  facilities layout problem, or it can treat them
  as a hierarchical set of cell problems.
• The approach supports both FLP and CM.

7
Genetic Algorithm
8
Genetic representation
Chromosome for single area
9
Genetic representation
Chromosome with hierarchical constraints
10
Placement Algorithm
11
Case Study

• 52 Machine tools
• 3408 complex components
• 734 part types
• Complex product structures
• Total distance travelled
• Direct distance 232Km
• Rectilinear distance 642Km

12
Random generation
13
Experimental Design
14
Hierarchy of areas
The number of generations was the only
significant factor.
Best configuration
15
Single area

• Significant factors
• Population size
• Probability of crossover
• Number of generations

Best configuration
16
Conclusions

• Developed a GA tool that can treat layouts as a
  single area or a hierarchy of cell layout
  problems.
• GA tool significantly better than random search
• GA worked better with unconstrained single area
  problems. In this case, population size,
  probability of crossover and number of
  generations were significant factors.
• With the hierarchy of cells approach only the
  number of generations was significant. Quality of
  layout influenced by initial allocation of
  machines to cells.