2006' 5' 24' Computer Aided Process Planning

1
2006. 5. 24. Computer Aided Process Planning
An Agent based optimization approach to
manufacturing process planning
Yonsei University Mechanical Engineering
Dept. Kim Yong woo
2
Contents

• Abstract
• Introduction
• Background
• Optimizing Agents
• Base Algorithm
• Modifications
• Agent based optimization algorithm
• Extension to continuous parameter problem
• Conclusion

3
Abstract

• Difficulty in optimization in manufacturing
  process planning
• - large number of potential configures
  (process sequence)
• - associated (process) parameters
• - space is highly discontinuous an multi-modal
• Agent based optimization algorithm combining
  stochatic techniques with knowledge based search

4
Introduction

• Gradient based strategies
• Simulated annealing
• Genetic algorithms

• Proposed approach
• - use agents within stochastic framework to
  configure and modify designs
• - the agents are a modification over the move
  sets typically used in simulated annealing

5
Background-problem definition

• Bulk manufacturing process planning
• Process planning problem provide description of
  the product in terms of its geometry and material
  data
• Two aspect to the problem
• 1. configuration problem
• - process sequence is to be selected out of
  mutitude of possible process sequence
• 2. instantiation problem the process parameters
  are to be optimized
• - focus on instantiation problem
• - consider in fixed manufacturing sequence

Casting -gt Upset forging -gt rough machining -gt
finish machining
6
Background previous approaches

• This research is in the category of generative
  process planning
• Process planning using agents on NC turning
  centers by Storr, Li and Stoehle(2000)
• Our emphasis is on the introduction of an
  optimizing agent-based algorithm with general
  application

7
Stoichastic Algorithms

• Genetic algorithm
• - maintain a population of designs
• Simulated annealing
• - initially large changes are made to the design
  state and towards the end only smaller changes
  are made
• A-Design
• - to use design agents to act on a
  representation of the problem domain to create
  design configurations, instantiations and
  modifications

8
Optimizing agents

• Agents
• - as perceiving their environment (the design
  state)
• - making judgments of how to effect change on ad
  design state
• - acting upon their environment through
  effectors (modifications to the design state)
• Combines process planning with optimization in an
  agent framework

9
Agent creation

• Three main issue in creation of agent
• - domain vs problem specific knowledge
• - deterministic vs stochastic selection
• - quantitative vs qualitative reasoning
• Domain knowledge is incorporated and agents are
  used in a stochastic framework
• For design optimization
• - identify the different aspect of design
• - make agents for each of them

10
Bulk manufacturing parameter optimizing agent

• Consider the bulk manufacturing process planning
  problem which involves a collection of
  manufacturing processes
• Agents are made corresponding to each process
• - change associated parameters
• - have domain knowledge about the process

11
Instantiation agents on research

• 15 instantiation agents
• 6 agents to decide the dimensions of the
  workpiece in each process
• (casting, upset forging, machine preform,
  blocker forging, closed die forging, rough
  machining)
• 9 agents for changing die speed, temp., friction
  factor in following process
• (upset forging, closed die forging, blocker
  forging)

12
Base algorithm

• Instantiation agent
• Intelligent search
• Faster convergence to optima
• Not production system
• Guideline for system
• Probabilistically make changes to a design

13
Base algorithm

• Given a sequence, the values of the parameter
  must be determined via instantiation agents
• Iterative strategy is similar to A-design
• - population of designs is maintained
• Genetic algorithms
• - the population as a whole progresses toward
  better and better solution

14
Result of base algorithm

• Evaluation is an important part of the design
  process
• The objective of the optimization algorithm is to
  find the manufacturing process design with the
  least possible cost

15
Modifications

• To improve base algorithm
• The probabilities of selection of the agents are
  kept constant throughout the iteration
• - dynamic information which might lead to
  faster convergence to optima is not taken

16
Modification 1

• Dynamically adjust probabilities of selecting
  each agents by a procedure based on the Hustin
  move set
• An analogy is formed between temperature and
  generation

17
Modification 1

• The initial probability of agents is directly
  proportional to the number of parameters it
  change
• The probability of selection of the ith agent is
  given by

• Result
• - the larger the quality factor for an agent the
  greater the probability that it will be applied
  at the next temperature

18
Modification 2

• Adjust the amount of perturbation to each design
  during modification
• - if more agents are called
• - then correspondingly more parameters are
  changed
• - thus increasing the possibility of the
  population member reaching the optimum faster

19
Modification 3

• Large changes are initially made to the design
  process parameters and progressively smaller
  changes are made as the iteration proceeds.

20
Modification 4

• Application of positive feedback
• - based on the assumption that agents which
  have been successful in the past will be
  successful in the future

21
Modification 5

• Application of negative feedback
• - assume that an agent that has been
  unsuccessful in the previous generations will
  have a higher likelihood of being unsuccessful in
  the future generations

22
Result of all modifications

• For sequence 1 and 3
• - The individual modifications have lower
  maximum, standard deviation and average compared
  to the base algorithm
• For sequence 2
• - the individual modifications have lower
  average and standard deviation

23
Agent based optimization algorithm

• Combines all the modifications into one algorithm
• Modification of Hustin move set for a Multi-agent
  scenario
• - if the agent-team has reduced the objective
  function value then they are rewarded in
  proportion of their previous probabilities

24
Complete Agent-Based optimization algorithm

• The population is generated randomly and sorted
• The number of agents to be called is decided by
  Modification 2, 3
• Objective function value of the new member is
  compared with the parent member to decide success
  or failure
• Depending on success or failure the agents
  probabilities are updated by Modification of
  Hustin set, positive, negative feedback

25
Complete Agent-Based optimization algorithm
lt Base algorithm gt
lt Complete agent-based optimization algorithm gt
26
Extension to continuous parameter problem

• Change the agent definition such that agents can
  select any real number within a given range
• Apply complete agent-based optimization algorithm
• The search space become considerably larger
• Standard deviation is higher to the discrete
  assumption

27
Conclusions

• Introduce agent-based optimization algorithm
• Combines deterministic and stochastic strategies
• Baser algorithm Modifications
• Complete agent-based optimization algorithm
• Future works
• - extension to process sequence
• - qualitative reasoning capability