Scheduler ProfileBased Intelligent Production Systems

1
Scheduler Profile-Based Intelligent Production
Systems

• Gürsel A. Süer
• IMSE, Ohio University, USA
• 7th International Workshop on Human Factors in
  Planning, Scheduling and Control in Manufacturing
  
• June 13-15, 2005, Groningen, Netherlands

2
Presentation Outline

• Manufacturing Systems
• Intelligent Production Systems
• Human Involvement
• Human Personalities
• Proposed Considerations
• Other Components

3
PART I. Manufacturing SystemsMajor Activities
4
PART I. Manufacturing Systems Suppliers,
Manufacturer, Customers
Materials Finished Products
Purchasing Production Planning Demand
and Control
Management Human Issues Human
Issues Human Issues
5
PART I. Manufacturing Systems Various Planning
Functions

6
PART I. Manufacturing Systems Characteristics of
Manufacturing Systems-I
Planning Horizon (t) Response Time
long
short
7
PART I. Manufacturing Systems Characteristics of
Manufacturing Systems-II
Precision
low
high
8
PART II. Intelligent Production
SystemsDefinitions

• Intelligent System
• Software than can make decisions to solve
  problems
• Intelligent Production Systems
• Software that contains the expertise required to
  make decisions and the tools needed to solve
  production-related problems
• In this presentation, the emphasis is on the
  production planning and control aspects

9
PART II. Intelligent Production Systems KB
Systems -Traditional Approach
10
PART II. Intelligent Production Systems What is
Expertise?

• Combination of
• Formal Knowledge
• Domain-independent knowledge
• Learning from schools and books
• Deep knowledge
• Experience
• Domain-dependent knowledge
• Learning from mentors and experiences
• Surface knowledge

11
PART II. Intelligent Production Systems Sample
Applications-I

• single machine scheduling advisor
• Experts Gürsel A. Süer and Cihan Dagli
• Developers Gürsel A. Süer and Cihan Dagli
• Users Whoever has the need for such a
• consultation

12
PART II. Intelligent Production Systems Sample
Applications-II

• Knowledge-Based Master Scheduling
• Expert Current Master Scheduler at
• AVON
• Developers Gürsel A. Süer
• User Current/Future Master Schedulers
• at AVON

13
PART II. Intelligent Production SystemsReplacing
Experts

• Knowledge-based systems can be developed
• to make the expertise available to others
• to speed up decision making process
• Eventually, experts are replaced by
  Knowledge-Based Systems

14
PART II. Intelligent Production Systems
Advantages of KB Systems

• 1. They are always available
• 2. They always attend to details
• 3. They are consistent
• 4. Better KB systems can perform their
  specialized
• tasks better than human specialists
• 5. They ask questions and explain their reasoning
  if
• asked and justify their questions
• 6. They can function with incomplete and
  uncertain
• data
• 7. They do not display biased judgments

15
PART II. Intelligent Production Systems
Disadvantages of KB Systems

• 1. They lack commonsense
• 2. Their expertise is limited to a narrow field
• 3. They do not learn
• 4. They cannot reason from axioms or general
• theories

16
PART II. Intelligent Production Systems Other
Approaches-I

• Fuzzy Sets
• Deals with Uncertainty
• Membership Function
• Neural Networks
• Extremely simplified models of human brain
• Different approach to information processing when
  an algorithmic procedure for solving a problem is
  not known
• Learning

17
PART II. Intelligent Production Systems Other
Approaches-II

• Tabu Search
• Neighborhood search
• Cycling back to previously visited solutions is
• prevented by the use of memories called tabu
  lists
• Efficient Technique
• Simulated Annealing
• Provides a means to escape local optima by
  allowing hill climbing moves
• Modeled after thermodynamic behavior

18
PART II. Intelligent Production Systems Other
Approaches-III

• Genetic Algorithms
• Bio-inspired technique
• Darwinian principle of natural selection
• Proved to be effective especially in the global
  search
• Memetic Algorithms
• Similar to Genetic Algorithms
• Exploits all available knowledge (heuristics,
  approximation algorithms, local search, local
  optimizers, etc.)
• Ant Colonies
• Behavior of real ants
• Pheromone trail laying and following

19
PART II. Intelligent Production Systems Other
Approaches-IV

• Particle Swarm Optimization
• social behavior of organisms such as bird
  flocking and fish schooling
• knowledge is optimized by social interaction
• population-based search procedure
• individuals change their position (state) with
  time
• each individual adjusts its position according to
  its own experience, and according to the
  experience of a neighboring individual

20
PART II. Intelligent Production Systems Other
Approaches-V

• Genetic Programming
• Extension of genetic algorithms
• Population consists of computer programs
• Data Mining
• Deals with analyze large databases to solve
  problems
• automated extraction information from databases

21
PART II. Intelligent Production Systems Other
Approaches-VI

• Scatter Search, Variable Neighborhood Search,
  Guided Local Search, Adaptive Search, Iterated
  Local Search, Constraint Satisfaction, etc.
• Classical Optimization
• Linear Programming
• Integer Programming
• Nonlinear Programming, etc.

22
PART II. Intelligent Production Systems From ..To
From Knowledge-Based Intelligent Systems
To Faster Systems Computation-Intensive
Systems More Efficient Search Techniques Learning
Systems Adaptive Systems Hybrid Techniques
23
PART III Human InvolvementIn a Typical Function
24
PART III Human Involvement In a Typical
Function -Details
25
PART III Human InvolvementDecision Making
Levels
Decisions
Solution Tech.
Decision Makers
Conscious
Opt.
Group
Eff. Search Fuzzy KB.
Unconscious
Single
26
PART III Human InvolvementIndividual vs. Group
DM

• Individual Decision Making
• Expertise (Knowledge and Experience)
• Behavior
• Group Decision Making
• Expertise of Individuals (Knowledge and
  Experience)
• Behavior of Individuals
• Group Dynamics Among Individuals

27
Part IV Human Personalities Elements of
Behavior

• Behavior depends on
• person
• environment
• anticipated future events (look ahead)

28
Part IV Human Personalities Classification of
Personal Types-I

• A. By Carl Gustav Jung
• Two dimensions
• 1. Attitudes (orientations)
• Introversion subjective experience
• Extraversion objective experience

29
Part IV Human Personalities Classification of
Personal Types-II

• 2. Functions
• Thinking intellectual function
• connect ideas
• understand nature of problem and then solve it
• Feeling pleasure
• anger
• pain
• love
• Sensing seeing
• hearing
• touching, etc.
• Intuiting hunch

30
Part IV Human Personalities Classification of
Personal Types-III

• Eight psychological types
• 1. Introversion thinking
• Emotionless, distant, arrogant, inconsiderate,
• stubborn, pursue their own thoughts
• 2. Extraversion thinking
• Impersonal, cold, repress feeling,
• objective reality is their driving force

31
Part IV Human Personalities Classification of
Personal Types-IV

• 3. Introversion feeling
• Feel intense emotions, keep them hidden, usually
  they have inner harmony but may erupt suddenly
• 4.Extraversion feeling
• Feelings change, emotional, moody, but sociable,
  
• intense but short-lived attachments
• 5. Introversion sensation
• Calm, self-controlled, boring,
• deemphasis on thoughts and feelings
• 6. Extraversion sensation
• Realistic, practical, hardheaded, accept world
  as is without giving much thought, sensation from
  their own experience

32
Part IV Human Personalities Classification of
Personal Types-V

• 7. Introversion Intuition
• Dreamers, visionaries, impractical. They may not
  communicate but others can benefit
• 8. Extraversion Intuition
• New worlds to conquer, good at promoting
• ideas, their interest is not sustained.

33
Part IV Human Personalities Classification of
Personal Types-VI

• B. Raymond Bernard Cattell
• Personality is a complex structure of various
  traits
• Started out with 4500 definitions
• Condensed the list to 200
• Later reduced it to 35 surface traits

34
Part IV Human Personalities Classification of
Personal Types-VII

• Classification of Traits
• 1. Surface traits, can be captured from
  observable behavior
• 2. Source traits, cannot be captured from
  observable behavior
• They can be determined by performing Factors
  Analysis. Surface traits are intercorrelated and
  factored in order to identify the influences that
  underlie them

35
Part IV Human Personalities Classification of
Personal Types-VIII

• Another Classification
• Dynamic traits (setting goal)
• Ability traits (effectiveness of reaching goal)
• Temperament traits (the way the person moves
  towards goal)

36
Part IV Human Personalities Classification of
Personal Types-IX Cattells 16 Personality Factors

• Outgoing Reserved
• More intelligent Less Intelligent
• Stable Emotional
• Assertive Humble
• Happy-go-lucky Sober
• Conscientious Expedient
• Bold Shy
• Tender minded Tough minded
• Suspicious Trusting
• Imaginative Practical
• Shrewd Forthright
• Apprehensive Placid
• Experimenting Traditional
• Self-sufficient Group-Tied
• Controlled Casual
• Tense Relaxed

37
Part IV Human Personalities Classification of
Personal Types-X

• C. Hans Jurgen Eysenck
• Three measures of personality dimensions
• 1. Introversion-Extraversion
• Introversion tender minded, seriousness,
  inhibited, performance interfered with by
  excitement, preference for solitary tasks, etc.
• Extraversion tough mindedness, impulsiveness,
  tendency to be outgoing, desire for novelty,
  performance enhanced by excitement, preference of
  tasks that require contact with other people,
  etc.

38
Part IV Human Personalities Classification of
Personal Types-XI

• 2. Neuroticism
• Below average emotional control, slowness in
• thought and action, lack of persistence, lack of
  
• sociability
• 3. Psychoticism
• Poor concentration, poor memory, insensitivity,
• lack of caring for others, cruelty, disregard
  for danger, occasionally originality or creativity

39
Part IV Human Personalities Classification of
Personal Types-XII
40
PART V Proposed ConsiderationsHuman Profiles-I

• There are various possibilities to incorporate
  human issues into intelligent production system
  development
• A. Human Profiles
• Typically, knowledge is extracted from an expert
  and KB system is built
• The rules established may represent his
• knowledge
• experience
• personality

41
PART V Proposed Considerations Human Profiles
-II

• The user is indirectly influenced by the Experts
  personality as well
• The idea is to keep
• knowledge SAME
• USE/IGNORE experience and
• VARY personality traits/strategies
• We need to define different Human Profiles and
  make them options during Consultation

42
PART V Proposed Considerations Human Profiles
-III

• Example
• A scheduler may be very optimistic in terms of
  availability of resources (i.e., no machine
  downtime, no material shortages, no worker
  absenteeism, no power failures, high efficiency,
  etc.)
• Another may be cautious and uses averages based
  on the past data whereas another may be very
  pessimistic and allows larger buffers in the
  schedule).
• You will get different schedules in each case.

43
PART V Proposed Considerations Human Profiles
-IV

• Example
• Another example may be processing and setup
  times. An optimistic scheduler uses standard
  times whereas a cautious scheduler may use
  slightly higher times and the pessimistic one may
  use high processing times.

44
PART V Proposed Considerations Human Profiles -V

• Example
• A scheduler may be in curious mode and ask more
  questions to get more accurate information to use
  in the planning process such as
  operator-operation times as opposed to only
  standard times.

45
PART V Proposed Considerations Human Profiles
-VI

• Example
• A scheduler may be in outgoing mode and gather
  information (formal/informal) about the
  availability of operators for a weekend overtime,
  in case needed, etc.

46
PART V Proposed Considerations Human Profiles
-VII

• Example
• A manager does not want to have more than 5 tardy
  jobs in that particular week. He runs the
  software in rigid mode and finds 7 tardy jobs
• Output from the software
• ? No such schedule exists
• No further action is taken.

47
PART V Proposed Considerations Human Profiles
-VIII

• Example
• However, the response may be as follows if
  software is run in thinking mode in the above
  problem
• Output from the software
• ? No such schedule exists
• ? However, if you allow overtime on Machine XX
  for 2 hours, you will reduce your number of
  tardy jobs to 5

48
PART V Proposed Considerations Human Profiles
-IX

• Example
• For the above example, even better approach would
  be to look into alternative actions and evaluate
  them in intelligent/analytical mode (overtime,
  subcontracting, alternative processes, additional
  machines, etc.)

49
PART V Proposed Considerations Human Profiles -X

• Example
• In the above example if the manager does not want
  to do overtime then it may be end of the
  consultation. However, if the user chooses
  persistence mode, software may generate other
  alternatives such as
• modify ready time
• ? if you bring in Material YY 2 days earlier,
  you can start processing W100 and W200 sooner and
  thus reduce number of tardy jobs to 5

50
PART V Proposed Considerations Human Profiles
-XI

• Example
• check if alternative materials are available
• In the above example, if material cannot be
  brought in early, then check the possibility if
  alternative materials can be used and also check
  their availability

51
PART V Proposed Considerations Human Profiles
-XII

• Example
• If the DM chooses to use bold option, it may
  perform a capacity analysis first and warns you
  that
• ? several tardy jobs are expected as capacity
  available is not sufficient, you will have to
  increase your capacity first!!

52
PART V Proposed Considerations Human Profiles
-XIII

• Example
• The software may run in aggressive and thinking
  mode where customers are classified,
  product-customer relations are established, then
  customers to be delivered are chosen.

53
PART V Proposed Considerations Human Profiles
-XIV

• Example
• The software may run in aggressive mode and
  suggest that you contact one of the customers and
  change due date or convince the customer for
  partial delivery to be able to reduce the number
  of tardy jobs to 5.

54
PART V Proposed Considerations Human Profiles
-XV

• Example
• The software may run in perfectionist mode and
  report that there is a schedule with 3 tardy jobs
  by running GA with higher population size and
  generations, BB, etc.
• Example
• The software may randomly choose overtime people
  in insensitive mode whereas it could skip elderly
  and women who has school-age children in a
  thoughtful mode.

55
PART V Proposed Considerations Human Profiles
-XVI

• Example
• The software may randomly choose overtime people
  in casual mode whereas it could only choose
  people with needed skills in controlled mode.
• Software may ask how much time is available to
  solve the problem (or rescheduling) and then
  chooses the best technique to use within time
  restrictions in self-sufficient mode

56
PART V Proposed Considerations Human Profiles
-XVII

• Example
• The software may generate a schedule where
  several products may have to be run every day to
  use the capacity better. However, in practical
  mode, small lots can be grouped and overtime may
  be suggested one day in the week.

57
PART V Proposed Considerations Human Profiles
-XVIII

• Example
• In no-risk mode, jobs with materials already in
  warehouse and firm orders will be scheduled
  whereas, in risky mode, jobs with anticipated
  material arrival times and anticipated customer
  orders can also be scheduled.

58
PART V Proposed Considerations Human Profile
Simulator

• B. Human Profile Simulator
• A manager may simulate the system by using
  different human profiles and eventually may
  choose one of the schedules
• He may sympathize with one of the profiles but my
  want to check others as well before he makes a
  final decision
• Different profiles may be more suitable for
  different occasions, periods, production units,
  shifts, etc.

59
PART V Proposed Considerations Human Profile
Evaluator

• C. Human Profile Evaluator
• It is good idea to keep track of past performance
  
• Correlation among past performance vs. chosen
  profiles should be determined
• An effort should be made to estimate the
  performance of other profiles in the same
  situations
• This information may be useful in deciding what
  profile to choose for the upcoming periods based
  on problem characteristics

60
PART V Proposed ConsiderationsPerformance
Measure Identifier-I

• D. Performance Measure Identifier
• Usually Decision Makers determine the performance
  measure to consider
• However, in many cases they follow tradition,
  habits, etc. without looking into problem
  specifics
• How would they know what measure is most relevant
  in this period without doing any extensive
  computation?

61
PART V Proposed Considerations Performance
Measure Identifier-II

• Ideally, software should make several iterations
  and passes to identify the critical performances
  to focus on
• We have made some efforts to solve a single
  machine scheduling problem by using Genetic
  Algorithms.
• GA runs with small population size and low
  generations. We identify critical measure(s) and
  run GA full version to improve that particular
  measure(s)

62
PART V Proposed Considerations Uncertainty-I

• E. Uncertainty
• Fuzzy is a good technique to deal with
  uncertainties
• Membership function contain subjective
  preferences
• Human Profiles can be incorporated into
  membership functions

63
PART V Proposed Considerations Uncertainty-II
µ
µ
1
1
0
0
nT
nT
5
8
12
5
DM X DM Y More tolerant Less tolerant
64
PART V Proposed Considerations Uncertainty-II
µ
µ
1
1
0
0
nT
nT
3
8
12
3
DM X DM Y More demanding More
demanding more tolerant less tolerant
65
PART V Proposed Considerations Robust Solutions

• F. Robust Solutions
• Robust solutions can be provided based on
  stochastic parameters (processing times, setup
  times, machine availability, etc.)
• performance measures
• acceptable solutions with respect to multiple
  performance measures
• combination of stochastic parameters and
  performance measures

66
PART V Proposed Considerations Multiple
Performance Measures-I

• G. Multiple Performance Measures
• Non dominated solutions
• Weighted approaches
• Lexicographical approach
• Fuzzy Genetic Algorithms can be used to deal
  with these problems as well
• GA fitness function includes what measures have
  been satisfied how well they have been
  satisfied
• How well is measured through satisfaction levels

67
PART V Proposed Considerations Multiple
Performance Measures-II
Various fitness functions can be used for
GA-based fuzzy scheduling FF min (?i) FF max
(?i) FF ?i (?i) FF ?i (ai ?i)
68
PART V Proposed Considerations Multiple
Performance Measures-II
Qualitative and Quantitative Measures can be
combined by using fuzzy as well Qualitative
evaluation can be done in interactive mode with
Decision Maker or by using membership
functions
69
PART V Proposed Considerations Dynamic
Environment-I

• H. Dynamic Environment
• Alternative membership functions are defined for
  different situations
• Situations may be environment-dependent and/or
  future dependent
• A company is more sensitive to reducing tardy
  jobs in a highly competitive period than in a
  less competitive period

70
PART V Proposed Considerations Dynamic
Environment-II

• A company is more willing to do overtime due to
  high demand than simply due to machine failure
• A company is reluctant to invest in new equipment
  when demand is flat. However, when trend is
  increasing demand, they will be less reluctant
• More suitable membership function is used in each
  instant
• Similarly, fuzzy rules can be revised

71
I PART V Proposed ConsiderationsInteractiactive
Operating Mode

• I. Interactive Operating Mode.
• The software should be able to evaluate any
  plan/scheduled suggested by human Decision Maker
  
• If the DM is satisfied with most of the schedule
  but asks a few changes, it should evaluate the
  impact of requested changes
• If the DM is satisfied with most of the schedule
  but still interested in improvement, neighborhood
  search can be conducted
• If the DM is satisfied with the schedule but now
  interested in a secondary performance measure,
  neighborhood search can be conducted

72
PART V Proposed Considerations Fuzzy Group
Decision Making-I

• J. Fuzzy Group Decision Making
• Individuals have their membership functions
• Membership functions may be similar or
  conflicting
• Individuals may have the same weight or varying
  weights
• If sum of satisfaction gt Th, schedule is
  acceptable
• GAs can be used to maximize the sum of
  satisfaction

73
PART V Proposed Considerations Fuzzy Group
Decision Making-II
If no schedule is found with satisfaction gt
Th discussions (simulation) start.
74
PART V Proposed Considerations Fuzzy Group
Decision Making-III

• Membership functions are modified to reflect
  discussions. Some personality profiles/traits
  determine whose membership functions can be
  modified and by how much. This process is
  repeated a number of times to obtain a
  resolution.
• No solution is a possibility as well
• Such a tool can be used to determine the makeup
  of a group so that working relations can be
  established

75
PART V Proposed Considerations Fuzzy Group
Decision Making-IV

• GAs can be used to solve these problem with
  Maxmin type of fitness function (Maximize the
  minimum satisfaction level)
• OR
• Maximize total satisfaction subject to an
  acceptable minimum satisfaction of each DM.
• (Max sum FF s.to. SLigtMIN)

76
PART V Proposed Considerations Fuzzy Group
Decision Making-V

• Another possibility is to have group discussion
  but single DM. Each person in the group is trying
  to influence the decision (membership function)
  of the Decision Maker.
• This problem can also be extended to Multi-DM and
  Multi-PM as well.

77
PART VI Other Components

• Distributed Systems
• Everybody connected and all relevant information
  available to the user
• Vision systems
• Status Check
• Evaluation of performance
• Mood Predictor

78
PART VI Other Components-II

• Natural Language Processing
• Automatic Instructions
• Computer vs. DM
• Computer vs. Implementers
• Automation
• Highly automated environment, robotics

79
PART VI Other Components-III

• Create Operator skill and preference database and
  use it during Task Allocation
• Extend these concepts to Supplier Evaluation
• Extend these concepts to Customer Evaluation