Frame-based architectures for manufacturing planning and control

Frame-based architectures for manufacturing planning and control

Artificial Intelligence in Engineering 7 (1992) 63-91 Frame-based architectures for manufacturing planning and control R. Karni & A. Gal-Tzur Faculty...

2MB Sizes 2 Downloads 62 Views

Artificial Intelligence in Engineering 7 (1992) 63-91

Frame-based architectures for manufacturing planning and control R. Karni & A. Gal-Tzur Faculty of bzdustrial Engineering and Management Technion - Israel Institute of Technology, Haifa, Israel 32000 (Received 24 October 1990; accepted 4 May 1991)

Effective manufacturing planning and control (MPC) necessitates coordination and integration of various aspects of demand, production and logistics management. A holistic approach is therefore the key to success in this field. A frame-based architecture should be ideally suited to constructing knowledge-based systems for MPC, as frames can represent entities in the planning process, rules can express interrelationships between these entities, and the planning strategy is paralleled by the inference procedure. Four applications are described in detail by means of four framebased paradigms: design of an operations regime; project planning of a new product launch: configuration of a process cell; and an analysis of the operation of an integrated manufacturing system. These architectures, and others presented in a previous article, are categorized as examples of generic tasks, a methodology proposed by Chandrasekaran-13 which defines underlying structures in terms of system goals, input/output characteristics, knowledge representation and inference strategy. The generic task approach appears to be useful in determining an appropriate architecture for a given MPC task, and also for designing and implementing the resultant knowledge-based system. industrial engineering, manufacturing planning and control, production and operations management, frame-based systems, generic tasks. Key words:

control of materials - - requires knowledge of several system impinging on it: marketing and sales, product design, process design, raw materials acquisition, transportation and performance tracking. Planning a project to bring a new product to market demands prior knowledge concerning the product to be developed and its associated production processes, the marketing environment, deadlines for decisions and deliveries, teams to be allocated to the project, and periods during the year when personnel and facilities may or may not be available. Selection of a processing configuration - machines and tools - - necessitates knowledge of product component attributes, processing operations on these components, and the capabilities, controllability and costs of available machines and tools. Finally, the operation of a manufacturing system is based upon coordination between strategy (business, marketing, production), planning (material, production and capacity scheduling), and execution (shop floor, purchasing and inventory). Frame-based AI architectures are ideally suited to MPC systems because of their integrative and interactive nature. Frames can represent entities in the planning process and their associated attributes. Rules can express the interrelationships between these entities. The inference engine can represent the strategy used to develop the plan or control strategy. In this article we detail four frame-based architectures

1 INTRODUCTION Engineering activity is oriented towards providing the right quantity, of the right product, to the right customer, at the right time, location, quality and price. The production or manufacturing function is central to achieving this overall purpose. It is often pointed out that effective implementation of this function necessitates coordination and integration of various aspects of demand, production and logistics management.~,2.8.~.lo 16,18.34 A holistic approach is the key to success in manufacturing planning and control; many of the difficulties associated with meeting production goals derive from a narrow focus on sub-areas of activity. A wider focus, however, requires extensive knowledge of these aspects and interactions between them. Burbidge, s'9'~° for example, in an exhaustive study of production system design, lists over 200 interrelated parameters which may be involved. The extent to which manufacturing planning and control (MPC) requires a wide range view is illustrated by some typical environments associated with basic activities in this area (see figures.). The design of an operations regime - - the flow and Art~fictal hnelhgence m Enguwering 0954-1810/92/$05 00

.c: 1992 Elsevier Science Pubhshers Ltd.

63

64

R. Karni, A. Gal-Tzur

to demonstrate how AI techniques can enhance the planning and design of manufacturing systems. These paradigms are presented in Tables 1 to 4, using a uniform scheme for describing each architecture and providing a concise, but complete, illustrative knowledge base. In each table, part (a) describes the knowledge and rule representation and the operation of the associated inference mechanism; part (b) presents the fact base: part (c) details the fact base: instance frames; and part (d) sets out the rule base.

[[I-o

[[I--I"

[[I--"I

OPERATIONS

Knowledge about co~nts diaensior~ and quantities

OPEIIATIOi~$

KnowledEe

about operatt~m

about

capabilities operations and Costa

and operation t~es

}

V

<

I

A

I about

I

for the product

)[

V I~flOUCT I

c~tz~llab£1tty

~intainability

I

Knowledge about the product V V for product [ i for source [ I ~l&~ a v a i l a b i l i t y ' FROOt~TION (_vailability, EXTRACTION I Knowledge Knowledge about Knowledge about sources needs and about production [ requirements

dellverab~lity

i

lity

Fig. 3. Schematic for manufacturingprocess configuration.

A

Knowledge

about t r a n s p o r t a t i o n

V

for product

., [ I ~ I ] ¢ l ~ q ~

performance

I Kn°wledge ab°ut

BUSItr~SS

CUSTCUER

Buslness planninl

Customer

performance

I

Knowledge

about

about

deliveries and priorities

I

f

I

I Knowledge

I

components

Knowledge about deadlines

and complexity

and complexity

i

about

I

I

'

Minster set,eddie

ACTIVITIES

I Knowledge about

Knowledge about types and avallabil£ties

I

Knowledge about events and time windows

Knowledge about markets and competltots [

1,

actual dates and produced quantities

I Shop ~lcor opetstion end ~t~l

Fig. 2. Schematic for projection planning (new product launch).

I

I

I

and required capacities

~qulra~t~ plann/ng t ~b~.dlng

I

I

Knowledge about scheduled dates

quantities

I Knowle~e about

actual dates and purchased quantities

I ~lsition operatlc~ cc~t~l

I

about scheduled dates

Kn~ledge about planned dates and batch/lot

~nd production quantities

~sourees

Knowledge about placated dates

[ f e a s i b l e dates

Knowledge abou t processes and

I

~te~tal ~iul~Jp-nte

about

V

1

1

Knowledge

needs and orders

l Knowledge about operations

and control

strategies and policies

about

I

planning

I

Knowledge

f e u i b l e dates and aggrel~te capacities

I

Production

plarmin~ and control

coatrol

Fig. 1. Schematic for production design.

Knowledge

Marketing

I

I Knowledge about actual inventories

I Inventory

I

I lead times and delivered quantities

I Supplier

operation

and control

Fig, 4. Schematic for manufacturingplanning and control.

Frame-based architectures for manufacturing planning and control

65

Table 1. A slot-driven planning system (a) Paradigm description (1) Planning environment The environment is described by a set of impact entities and their association characteristics. The plan is described as a separate entity, which is developed on the basis of these characteristics by a straight-forward sequential process of obtaining or deriving all slot values related to the environment and the plan. (2) Entity representation (fact base) Each impact entity and the plan itself are single-instance frames. Impact slot values are usually determined by interaction with the planner ('ASKFOR'), Plan slot values are determined by slot-specific domain operators ('ATTACH') using inheritance to reference impact frames. (3) Problem-solving domain operators (rule base) Domain operators have the form of situation-action I F . . . THEN rules, partitioned into sets attached to a specific slot ('SET'). Their function is to determine a value for the slot. (4) User interface (fact base, rule base, inference engine) The planner can be requested ('ASKFOR') to indicate a slot value from a predetermined list ('VALUE') or enter a value for the slot. (5) Control operators (inference engine) 0) General sequence The inference mechanism carries out a single pass over all frames and their slots, from top-to-bottom in the order in which they are defined in the fact base ('INSTANCES'). It determines a value for each slot by requesting it from the user ('ASKFOR'), using domain operators ('ATTACH'), or, by failing these, using a defaulted value ('PRESET'). (ii) Rule set sequence As each slot is referenced, the inference mechanism carries out multiple iterations over the associated set of rules ('SET'). At each iteration, rules are tested from top-to-bottom in the order in which they are defined in the knowledge base. As each rule is tested, it may be skipped (a parameter value will only be determined in a subsequent iteration), confirmed or rejected. Rules which have been confirmed or rejected are ignored during subsequent iterations. Whenever a rule is confirmed and its consequent section activated, the iteration is terminated and testing for the subsequent iteration begins with the first remaining rule m the rule set. Iteration continues until the value of the slot has been determined. If no value can be determined by rule activation, the defaulted ('PRESET') value is inserted. (iii) Termination Termination occurs when values have been allocated to all slots.

Table 1. A knowledge base for production design (cont.) (b) Knowledge base: class frames *CLASSES

FRAME SLOT

PRODUCTS VARIETY

SLOT

TREE

SLOT

MODELS

SLOT

TYPE

SLOT SLOT

NOVEL PRICE

LITERAL VALUE VALUE VALUE LITERAL VALUE VALUE LITERAL VALUE VALUE VALUE LITERAL VALUE VALUE YES/NO LITERAL VALUE VALUE

ASKFOR SINGLE SEVERAL MANY ASKFOR SIMPLE COMPLEX ASKFOR SINGLE FEW MANY ASKFOR REGULAR SPECIAL ASKFOR ASKFOR LOW HIGH

LITERAL VALUE VALUE VALUE VALUE

ASKFOR LOCAL COUNTRY OVERSEAS WORLD

Product/mix characteristics Number of products produced Single product type Several products Wide variety of products Complexity of structure (tree) Simple product Complex product Options/features offered Single model A few models offered Many models offered Product standardization Regular (standard) product Customized product Product novelty (newness) Price of each item Low-priced items High-priced items

Jg

FRAME SLOT

MARKET LOCATION

Product market characteristics Market distribution Only marketed locally Marketed country-wide Only marketed overseas Marketed country and overseas

R. Karni, A. Gal-Tzur

66

SLOT

VOLUME

SLOT

ORDERS

SLOT

LEADTIME

SLOT

CUSTOMER

SLOT

OUTLETS

SLOT

DEMAND

SLOT

COMPETE

SLOT

FORECAST

FRAME SLOT

SUPPLIES PLACE

SLOT

SUPPLIER

SLOT

STOCK

SLOT

STANDARD

FRAME SLOT

PROCESS TECHNO

SLOT

MACHINE

SLOT

MANPOWER

FRAME SLOT

REGIME CELLTYPE

SLOT

SHIFTS

SLOT

FINISHED

LITERAL VALUE VALUE VALUE LITERAL VALUE VALUE VALUE LITERAL VALUE VALUE VALUE LITERAL VALUE VALUE VALUE LITERAL VALUE VALUE VALUE VALUE LITERAL VALUE VALUE VALUE VALUE LITERAL VALUE VALUE VALUE VALUE VALUE YES/NO

ASKFOR SMALL MEDIUM HEAVY ASKFOR REGULAR PHONE SPECIAL ASKFOR FAST MEDIUM LONG ASKFOR PRIVATE INDUSTRY MILITARY ASKFOR FACTORY BRANCH WHOLE RETAIL ASKFOR FLAT SEASONAL UPTREND SPECIAL ASKFOR MONOPOLY NOVELTY SPECIAL PRICE DELIVERY ASKFOR

Demand (sales) volume Small volume medium volume Large volume Ordering characteristics Regular long-term orders Many phoned-in orders Few special orders Delivery lead time Fast dehvery times Medium delivery times Long delivery times Customer classification Private consumption Industrial consumption Military consumpUon Sales points for products Sales at factory Sales at company branches Sales to wholesalers Sales to retail stores Demand profile Horizontal Seasonal Upward trend Specialized orders only Nature of competition No competition A new product A customized product A low-priced product Fast delivery times Ability to forecast demand

LITERAL VALUE VALUE VALUE LITERAL VALUE VALUE LITERAL VALUE VALUE VALUE YES/NO

ASKFOR LOCAL OVERSEAS WORLD ASKFOR MONOPOLY MANY ASKFOR LOW MEDIUM HIGH ASKFOR

Supplier/raw materials attributes Supplier location(s) Local supplier(s) only Overseas supplier(s) only Local and overseas supplier(s) Flexibility of supply A single supplier Many suppliers Cost of maintaining r/m stock Low r/m mventory cost Medium r/m inventory cost High r/m inventory cost Standard raw material items

LITERAL VALUE VALUE VALUE LITERAL VALUE VALUE VALUE LITERAL VALUE VALUE

ASKFOR REGULAR SPECIAL NEW ASKFOR SINGLE SEVERAL MANY ASKFOR SIMPLE SKILLED

Production process characteristics Reliability of technology Regular (standard) technology Specialized technology Innovative technology Number of machines/processes A smgle process Several processes Many processes Manpower qualifications Simple skills required Complex skills reqmred

LITERAL VALUE VALUE VALUE VALUE LITERAL VALUE VALUE VALUE LITERAL

ATTACH JOBSHOP FLOWSHOP GROUPED FLOWLINE ATTACH NEVER SEASONAL ALWAYS ATTACH

Production process characteristics Production cell organization Jobshop (mixed batch) Flowshop (mixed flow) Group technology Flow line (uniform batch) Necessity for shifts/overttme No shifts/over time required Seasonal shiftwork required Shifts/overtime always needed Product inventory control

VALUE VALUE VALUE

QUANTITY REVIEW ORDER

Make to stock (fixed OQ) Make up to stock (fixed review) Make to order

Frame-based architectures for manufacturing planning and control

SLOT

SCHEDULE

SLOT

LOTSIZE

SLOT

SAFETY

SLOT

DELIVERY

LITERAL VALUE VALUE VALUE VALUE LITERAL VALUE VALUE VALUE LITERAL VALUE VALUE VALUE LITERAL VALUE VALUE VALUE

ATTACH OP/OQ MRP JIT OPT ATTACH FIXED LOT4LOT OPTIMAL ATTACH LOW MEDIUM HIGH ATTACH FACTORY CONTRACT CLIENT

Scheduling methodology Order point and order quantity Material requirements planning Just in time (kanban) Optimal production technology Size of production lots Fixed lot size One lot per demand order Lot sizes optimized Safety stock required Low safety stock required Medium safety stock required High safety stock required Transportation of products By factory vehicles By sub-contracted trucker By client (customer) vehicles

(c) Knowledge base: instanceframes *INSTANCES , FRAME FRAME FRAME FRAME FRAME

ITEMS SALES VENDORS SHOP POLICY

PRODUCTS MARKET SUPPLIES PROCESS REGIME PARENT PARENT PARENT PARENT

ITEMS SALES VENDORS SHOP

Product characteristics Market characteristics Raw material characteristics Shop floor characteristics Production characteristics

(d) Knowledge base: rules

SET CELLTYPE IF AND AND AND OR THEN

TREE TECHNO VOLUME VARIETY MODELS CELLTYPE

EQ EQ NE EQ EQ EQ

SIMPLE REGULAR HEAVY SEVERAL FEW FLOWSHOP

IF AND OR AND AND THEN

TREE MACHINE MANPOWER VOLUME VARIETY CELLTYPE

EQ EQ EQ NE NE EQ

COMPLEX MANY SKILLED SMALL SINGLE GROUPED

IF AND AND AND AND THEN

MANPOWER VOLUME TYPE PRICE COMPETE CELLTYPE

EQ EQ EQ EQ EQ EQ

SIMPLE HEAVY REGULAR LOW PRICE FLOWLINE

IF AND AND AND AND THEN

MANPOWER VOLUME TYPE LEADTIME COMPETE CELLTYPE

EQ EQ EQ EQ EQ EQ

SIMPLE HEAVY REGULAR FAST DELIVERY FLOWLINE

LET

CELLTYPE

EQ

JOBSHOP

SET

SHIFTS

IF AND THEN

DEMAND LEADTIME SHIFTS

EQ EQ EQ

SEASONAL FAST SEASONAL

67

68

R. Karni, A. Gal-Tzur

IF AND THEN

VOLUME FORECAST SHIFTS

EQ EQ EQ

MEDIUM NO SEASONAL

IF AND THEN

VOLUME LEADTIME SHIFTS

EQ EQ EQ

HEAVY FAST ALWAYS

IF AND THEN

TECHNO MANPOWER SHIFTS

EQ EQ EQ

NEW SIMPLE ALWAYS

IF OR THEN

DEMAND CUSTOMER SHIFTS

EQ EQ EQ

UPTREND MILITARY ALWAYS

LET

SHIFTS

EQ

NEVER

SET

FINISHED

IF OR OR OR OR OR THEN

MODELS PRICE FORECAST NOVEL ORDERS COMPETE FINISHED

EQ EQ EQ EQ EQ EQ EQ

MANY HIGH NO YES SPECIAL SPECIAL ORDER

IF AND AND OR THEN

MODELS FORECAST LEADTIME CUSTOMER FINISHED

NE EQ EQ EQ EQ

MANY YES FAST INDUSTRY REVIEW

LET

FINISHED

EQ

QUANTITY

SET

SCHEDULE

IF OR OR OR OR THEN

MODELS FORECAST NOVEL ORDERS TREE SCHEDULE

EQ EQ EQ EQ EQ EQ

MANY NO YES SPECIAL COMPLEX MRP

IF AND AND AND THEN IF AND OR THEN

MODELS FORECAST LEADTIME MACHINE SCHEDULE VARIETY ORDERS ORDERS SCHEDULE

NE EQ EQ NE EQ NE EQ EQ EQ

MANY YES FAST MANY JIT MANY SPECIAL PHONE OPT

LET

SCHEDULE

EQ

OP/OQ

SET

LOTSIZE

IF OR OR OR AND THEN

CUSTOMER DEMAND PRICE NOVEL SCHEDULE LOTSIZE

NE EQ EQ NE EQ

PRIVATE SPECIAL HIGH YES OP/OQ LOT4LOT

IF OR OR OR AND THEN

COMPETE SCHEDULE TREE PRICE DEMAND LOTSIZE

EQ EQ EQ EQ NE EQ

PRICE MRP COMPLEX HIGH SPECIAL OPTIMAL

LET

LOTSIZE

EQ

FIXED

SET

SAFETY

EQ

Frame-based architectures for manufacturing planning and control

IF AND AND AND OR THEN

SCHEDULE PRICE STOCK LOCATION COMPETE SAFETY

EQ NE NE NE NE EQ

OP/OQ HIGH HIGH LOCAL MONOPOLY HIGH

IF AND AND THEN

SCHEDULE ORDERS LEADTIME SAFETY

EQ NE NE EQ

OP/OQ SPECIAL FAST MEDIUM

LET

SAFETY

EQ

LOW

SET

DELIVERY

SET OR OR AND THEN

LOCATION LOCATION CUSTOMER DEMAND DELIVERY

EQ EQ EQ EQ EQ

OVERSEAS WORLD MILITARY SPECIAL CLIENT

IF AND THEN

VOLUME ORDERS DELIVERY

NE NE EQ

LARGE REGULAR CONTRACT

LET

DELIVERY

EQ

FACTORY

69

*END

Table 2. A rule-driven planning system

(a) Paradigmdescription (1) Planning environment The environment is described by a set of impact entities and their associated characteristics. The plan is described by a set of sibling instances as a series of actions. It is developed on the basis of the impact characteristics by an iterative process of obtaining or deriving, and possibly modifying, all slot values related to the plan. (2) Enuty representation (fact base) Frame classes are prototypes of impacts or plan actions. Impact entities are single- or multiple-instance frames. The plan itself is made up of multipleinstance frames. Planning scenario slot values are determined by interaction with the planner ('ASKFOR'); plan slot values are developed by a domain operators ('ATTACH') using inheritance to reference other frames. (3) Problem-solving domain operators (rule base) Domain operators have the form of situation-action IF . . . THEN rules, partiUoned into sets for carrymg out specified actions ('SET'). Each set may apply to a specific frame mstance, or to all the instances of a specific frame class ('By . . .'). Its function is to determine value for slots referenced in the rule set. (Some slots, with default values ('PRESET'), may not be referenced in the rule base.) (4) User interface (fact base, rule base, inference engine) The planner can be requested ('ASKFOR') to indicate a slot value from a predetermined list ('VALUE') or enter a value for the slot. (5) Control operators (inference engine) (i) General sequence The reference mechanism carries out multiple passes over all rule sets, beginning from the 'START' rule set. Each set is terminated by a 'TO' operator, which switches the scan to the rule set named. Routing may also be specified by the 'RULESET' argument. Each rule set references either a umque single frame instance (no explicit reference given) or all multiple instances of a unique frame class ('BY . . .'). In some cases, all instances of one class may be referenced against all instances of the same or another class ( ' B Y . . . B Y . . . ' ) . If the same class is referred to a second time, the slots of the second reference are differentiated by the prefix '%'. (ii) Rule set sequence The inference mechamsm carries out multiple iterations over each set of rules for each frame instance referenced. At each iteration, rules are tested from top-to-bottom in the order in which they are defined in the knowledge base, and confirmed or rejected. Rules which have been confirmed are ignored during subsequent iterations. Whenever a rule is confirmed and its consequent section activated, the iteration is terminated. Testing for the subsequent iteration begins with the first remaining rule in the rule set and continues until no rule can be confirmed.

70

R. Karni, A. Gal-Tzur

(ui) Termination Termination occurs when the 'QUIT; ruleset is encountered.

Table 2. A knowledge base for project management (cont.) (b) Knowledge base: class frames * CLASSES FRAME SLOT

GLOBAL PROCESS

SLOT

PRODUCT

SLOT SLOT SLOT

COMPETE COMMENCE COMPLETE

LITERAL VALUE VALUE VALUE LITERAL VALUE VALUE VALUE YES/NO NUMERIC NUMERIC

ASKFOR SIMPLE MEDIUM COMPLEX ASKFOR SIMPLE MEDIUM COMPLEX ASKFOR ASKFOR ATTACH

New product planning scenario Process (technology) complexity Fairly simple process Somewhat complex process Highly complex process Product structure complexity Fairly simple structure Somewhat complex structure Highly complex structure Expected presence of competition Planned project start date Planned project completion date

FRAME SLOT SLOT

WINDOW W-BEGIN W-FINISH

LITERAL LITERAL

PRESET PRESET

Time window constraint Window start date Window end date

FRAME SLOT SLOT SLOT

CAUSAL FACTOR- 1 FACTOR-2 FACTOR-3

NUMERIC NUMERIC NUMERIC

ASKFOR ASKFOR ASKFOR

Causal Impact Impact Impact

FRAME SLOT SLOT SLOT SLOT SLOT SLOT

RESOURCE DAILY LEVEL ADD-ON R-BEGIN R-FINISH INACTIVE

NUMERIC NUMERIC FRAME LITERAL LITERAL FRAME VALUE VALUE

PRESET PRESET PRESET ASKFOR ASKFOR ATTACH VACATION WORKSHOP

Project resource type (Maximum) daily availability (Default) intensity per activity CAUSAL constraint on duration Earliest availability Latest availability Time WINDOW when resource inactive Unavailability due to vacation Unavailabdity due to training

FRAME SLOT ,

MILESTON MILEDATE

LITERAL

PRESET

Milestone constraint or goal Target date to achieve milestone

FRAME SLOT

ACTIVITY PHASE

SLOT SLOT

TARGET RESTYPE

SLOT SLOT SLOT SLOT SLOT SLOT SLOT SLOT SLOT SLOT

INTENSE INPUT OUTPUT ENABLE DISABLE CONTENT DURATION PRECEDE LOGICAL PRIORITY

SLOT SLOT

BEGIN FINISH

LITERAL VALUE VALUE VALUE VALUE VALUE FRAME FRAME VALUE VALUE VALUE NUMERIC LITERAL LITERAL YES/NO YES/NO NUMERIC NUMERIC FRAME FRAME NUMERIC VALUE VALUE VALUE NUMERIC NUMERIC

PRESET INITIAL COMMIT PLANNING PREPARE KICKOFF PRESET PRESET DESIGN PROCESS MARKET PRESET PRESET PRESET PRESET PRESET ATTACH ATTACH ATTACH ATTACH ATTACH 1 2 4 ATTACH ATTACH

Project activity identifier Phase in planning and design Initial study phase Determination of basic facts Specification of basic plans Preparation for production Commencement of production/sales Milestone target (MILESTON class) Resource type (RESOURCE class) Product design group Process design group Sales/marketing group Intensity (man-days per day) Required input to activate Required output Enabling condition ('crashable') Disabling condition (unavailable) Work content (man-days) Time duration (Typical) technological predecessor (Typical) logical predecessor Conflict priority rating Low expediting priority Medium expediting priority High expediting priority Planned start date Planned completion date

constraint on activities (add-on or multiply) factor (add-on or multiply) factor (add-on or multiply) factor

Frame-based architectures for manufacturing planning and control Table 2. A knowledge base for project management (cont.)

(c) Knowledgebase: instanceframes *INSTANCES Project planning scenario

FRAME

SCENARIO

GLOBAL

FRAME SLOT SLOT

VACATION W-BEGIN W-FINISH

WINDOW PRESET PRESET

24 26

FRAME SLOT SLOT

WORKSHOP W-BEGIN W-FINISH

WINDOW PRESET PRESET

18 19

FRAME SLOT SLOT SLOT

TREE FACTOR- 1 FACTOR-2 FACTOR-3

CAUSAL PRESET PRESET PRESET

1.00 1.15 1.50

Product complexity time increase Multiplier for simple product Multiplier for medium product Multiplier for complex product

FRAME SLOT SLOT SLOT

MAKE FACTOR- 1 FACTOR-2 FACTOR-3

CAUSAL PRESET PRESET PRESET

0.85 1.00 1.45

Process complexity time increase Multiplier for simple process Multiplier for medium process Multiplier for complex process

FRAME SLOT SLOT SLOT

MARKET DAILY LEVEL INACTIVE

RESOURCE PRESET PRESET PRESET

4 2 TRAINING

FRAME SLOT SLOT SLOT SLOT

DESIGN DAILY LEVEL ADD-ON INACTIVE

RESOURCE PRESET PRESET PRESET PRESET

5 1 TREE VACATION

FRAME SLOT SLOT SLOT

PROCESS DAILY LEVEL ADD-ON

RESOURCE PRESET PRESET PRESET

3 1 MAKE

FRAME SLOT

FEASIBLE MILEDATE

MILESTON PRESET

15

FRAME SLOT SLOT SLOT SLOT SLOT SLOT SLOT

P-DESIGN PHASE RESTYPE INTENSE OUTPUT DISABLE DURATION PRECEDE

ACTIVITY PRESET PRESET PRESET PRESET PRESET PRESET PRESET

PLANNING DESIGN 1 SPECIFY YES 4 PRODTYPE

FRAME SLOT SLOT SLOT SLOT SLOT SLOT SLOT

P-SALES PHASE RESTYPE INTENSE INPUT ENABLE DISABLE CONTENT

ACTIVITY PRESET PRESET PRESET PRESET PRESET PRESET PRESET

PREPARE MARKET 2 SALEPLAN YES YES 6

FRAME SLOT SLOT SLOT SLOT SLOT SLOT SLOT SLOT SLOT

METHODS PHASE RESTYPE INTENSE INPUT OUTPUT ENABLE DISABLE DURATION PRECEDE

ACTIVITY PRESET PRESET PRESET PRESET PRESET PRESET PRESET PRESET PRESET

PLANNING PROCESS 1 PROSTUDY PROPLAN YES NO 3 PROCNEED

Summer vacation interval

Marketing strategy workshop

Sales/marketing group

Product design group

Process/production design group

Decision to proceed with project Target date to reach a decision Product design in principle

Principal design specification

Study alternative solutions Prepare for sales (marketing)

Specify production regime

Process plan

Determine process requirements

71

R. Karni, A. Gal-Tzur

72

Table 2. A knowlege base for project management (cont.) (d) Knowledge base: rules *RULES * Set planning scenario SET LET LET LET LET LET END

START PRODUCT PROCESS COMPETE COMMENCE CHANGE START

BY SCENARIO .9 ? ? 9 EQ 0 TO ALLOCATE

* Allocate resource and time requirements SET IF THEN IF AND THEN

ALLOCATE INTENSE INTENSE DURATION CONTENT DURATION

BY EQ EQ EQ NE EQ

IF AND THEN END

CONTENT DURATION CONTENT ALLOCATE

EQ NE EQ TO

ACTIVITY 0 LEVEL 0 0 CONTENT/INTENSE 0 0 DURATION * INTENSE PREFER

PREFER INTENSE PRIORITY COMPETE PHASE PRIORITY PRIORITY PREFER

BY GT EQ EQ EQ EQ EQ TO

ACTIVITY LEVEL 4 YES INITIAL 2 1 ADJUST

ADJUST RESTYPE COMPETE INTENSE INTENSE DURATION RESTYPE PRODUCT DURATION CONTENT RESTYPE PRODUCT DURATION CONTENT RESTYPE PRODUCT DURATION CONTENT ADJUST

BY EQ EQ LT EQ EQ EQ EQ EQ EQ EQ EQ EQ EQ EQ EQ EQ EQ TO

ACTIVITY MARKET YES DAILY INTENSE + 1 CONTENT/INTENSE DESIGN SIMPLE DURATION * FACTOR-1 DURATION * INTENSE DESIGN MEDIUM DURATION * FACTOR-2 DURATION * INTENSE DESIGN COMPLEX DURATION * FACTOR-3 DURATION * INTENSE LOGIC

LOGIC INPUT %OUTPUT LOGICAL LOGIC

BY NE EQ EQ TO

ACTIVITY NULL INPUT %ACTIVITY RESET

BY EQ EQ EQ EQ TO

ACTIVITY COMMENCE COMMENCE + DURATION FINISH - 1 COMPLETE ~" FINISH FORWARD

* Set activity priority levels SET IF THEN IF AND THEN LET END

* Adjust durations by impact factors SET IF AND AND THEN IF AND THEN IF AND THEN IF AND THEN END * Insert a logical predecessor SET IF AND THEN END

* Initialize activity begin/fimsh dates SET LET LET LET LET END

RESET BEGIN FINISH FINISH COMPLETE RESET

Frame-based architectures for manufacturing planning and control

* Forward pass CPM iteration SET IF AND THEN

IF AND THEN

END

FORWARD PRECEDE BEGIN BEGIN FINISH FINISH COMPLETE CHANGE LOGICAL BEGIN BEGIN FINISH FINISH COMPLETE CHANGE FORWARD

BY NE LT EQ EQ EQ EQ EQ NE LT EQ EQ EQ EQ EQ TO

ACTIVITY NULL %FINISH + 1 %FINISH + 1 BEGIN + DURATION FINISH - 1 COMPLETE ~, FINISH 1 NULL %FINISH + 1 %FINISH BEGIN + DURATION FINISH - 1 COMPLETE ~, FINISH 1 CONVERGE

* Check for convergence of the forward pass SET IF THEN IF THEN END

CONVERGE CHANGE RULSET CHANGE CHANGE CONVERGE

EQ0 EQ CONFLICT EQ 1 EQ0 TO FORWARD

* Resolve a sequencing conflict by priority SET IF AND AND AND

CONFLICT RESTYPE PRIORITY %BEGIN %FINISH

BY NE GE LE GE

ACTIVITY %RESTYPE %PRIORITY FINISH BEGIN

THEN

%BEGIN % FINISH %FINISH COMPLETE CHANGE CONFLICT

EQ EQ EQ EQ EQ TO

FINISH + 1 %BEGIN + % DURATION FINISH - 1 COMPLETE ~ %FINISH 1 SHIFT

BY EQ NE GE LE EQ EQ EQ EQ EQ TO

ACTIVITY YES NULL W-BEGIN W-FINISH W-FINISH + 1 BEGIN + DURATION FINISH - 1 COMPLETE ~. FINISH 1 CHANGED

EQ EQ EQ TO

1 0 FORWARD TARGET-1

BY EQ EQ EQ GT EQ EQ TO

ACTIVITY YES FEASIBLE YES MILEDATE NO 1 TARGET-2

END

* Shift activities disabled during time window SET IF AND AND AND THEN

END

SHIFT DISABLE INACTIVE FINISH BEGIN BEGIN FINISH FINISH COMPLETE CHANGE SHIFT

* Test whether constraints have changed planned dates SET IF THEN END

CHANGED CHANGE CHANGE RULESET CHANGED

* If milestone not achieved cancel resource unavailability SET IF AND AND AND THEN END

TARGET- 1 ENABLE TARGET DISABLE FINISH DISABLE CHANGE TARGET-1

73

R. Karni, A. Gal-Tzur

74

* If milestone not achieved increase resource commitment SET IF AND AND AND AND THEN

END

TARGET-2 CHANGE ENABLE TARGET FINISH INTENSE INTENSE DURATION CHANGE;EQ 1 TARGET-2

BY ACTIVITY EQ 0 EQ YES EQ FEASIBLE GT MILEDATE LT DAILY EQ INTENSE + I EQ CONTENT/INTENSE TO ALTERED

* Test whether constraints have changed planned dates SET IF THEN END

ALTERED CHANGE CHANGE RULESET ALTERED

EQ EQ EQ TO

1 0 RESET ENDOFF

* To terminate and display project dates SET LET LET END

ENDOFF COMMENCE COMPLETE ENDOFF

--TO QUIT

*END

Table 3. A hypothesis-formation-driven planning system (a) Paradigmdescrip~on (1) Planning environment The environment is described by two sets of entities: (a) a set of predefined candidate plan components and their associated characteristics; and (b) a set of predefined reqmrements to be covered or fulfilled by the plan components. The plan developed is that subset of components which (a) covers all the requirements ('coverage goal'), (b) minimizes the complexity of the plan ('parsimony goal;) and (c) maximizes the desirability of the plan ('preference goal'). (2) Entity representation (fact base) Candidate plan components and plan requirements are defined by multiple-instance sibling frames. All slot values characterizing these components are given

a-priori ('PRESET'), or may be input or overridden if necessary ('ASKFOR'). A global entity contains general and scenario parameters. (3) Problem-solving domain operators (rule base) Domain operators have the form of LET (compute) rules or situation-action IF . . . THEN rules, partitioned into sets ('SET') for carrying out specific key tasks during the action of the planning algorithm. These include: 'SCENARIO' for indicating which requirements are to be considered when developing a plan; 'ACCEPT' for specifying acceptability of partial plans and candidate components, 'PRUNE' for eliminating partial plans that violate certain overall limitations; and 'PREFER' for ranking complete plans in preference order. (4) User interface (fact base, inference engine) The planner is requested ('ASKFOR') to indicate scenario values, such as plan acceptability limits. (e) Control operators (inference engine) (i) General sequence In response to each successive requirement in the 'SCENARIO' rule set, the inference mechanism activates a 'composite hypothesis formation' mechanism 24, 33, and finds and maintains a list of different subsets of candidates, or 'competing plans', which fulfil the first two goals ('coverage' or 'parsimony'). When all requirements have been elicited, it determines that plan, amongst the competing plans, which best fulfils the third goal ('desirability'). (i0 Rule set sequence As the planning algorithm proceeds, the inference mechanism carries out a single iteration over each associated set of rules, from top-to-bottom in the order in which they are defined in the knowledge base.

Frame-based architectures for manufacturing planning and control

75

(iii) Termination Termination occurs when all requirements have been specified and a preferred plan has been selected. (iv) The composite hypothesis formation model (a) Initially, the set of 'competing plans' is empty. (b) Get the next requirement from the scenario list. (c) Scan the set of competing plans. For each plan, check whether at least one of its components meets that requirement ('ACCEPT'). If not, search for an acceptable component ('ACCEPT') and include it in the plan such that the new plan fulfils the requirement set. If there are several such components, generate several corresponding competing plans. (d) Scan the set of competing plans. For each plan, check whether it meets overall constraint limitations ('PRUNE'). If not, delete the plan from the set. (e) Repeat these steps until all requirements have been specified and covered. (0 If more than one competing plan remains, selected the most preferred plan on the basis of a preference ranking ('PREFER').

Table 3. A knowledge base for cell configuration (cont.)

(b) Knowledgebase: classframes *CLASSES FRAME SLOT SLOT SLOT SLOT SLOT SLOT FRAME SLOT

GLOBAL MAXCOST MAXSUPP MAXCONT VENDORS CONTROLS COST MACHINE MT-TYPE

SLOT

SUPPLIER

SLOT SLOT

PRICE CONTROL

SLOT SLOT SLOT SLOT SLOT SLOT FRAME SLOT

TABLE-X TABLE-Y TABLE-Z AXES ACCURACY POWER PROCESS PRO-TYPE

FRAME SLOT SLOT SLOT SLOT SLOT FRAME SLOT

PRODUCT SHAPE-X SHAPE-Y SHAPE-Z FREEDOM EXACT OPERATE OP-TYPE

SLOT

ENERGY

NUMERIC NUMERIC NUMERIC NUMERIC NUMERIC NUMERIC

ASKFOR ASKFOR ASKFOR ATTACH ATTACH ATTACH

LITERAL VALUE VALUE VALUE VALUE LITERAL VALUE VALUE NUMERIC LITERAL VALUE VALUE VALUE

ASKFOR CENTER LATHE MILL DRILL ASKFOR MILICRON INGERSOL ASKFOR ASKFOR FANUK SIEMENS GE

VALUE NUMERIC NUMERIC NUMERIC NUMERIC NUMERIC NUMERIC

IBH ASKFOR ASKFOR ASKFOR ASKFOR ASKFOR ASKFOR

LITERAL VALUE VALUE VALUE VALUE VALUE VALUE VALUE

ASKFOR TURNING MILLING DRILLIMG INTHREAD OUTHREAD ECTHREAD CUTTING

NUMERIC NUMERIC NUMERIC NUMERIC NUMERIC

ASKFOR ASKFOR ASKFOR ASKFOR ASKFOR

LITERAL VALUE VALUE VALUE VALUE VALUE VALUE VALUE NUMERIC

ASKFOR TURNING MILLING DRILLING INTHREAD OUTHREAD ECCTHREAD CUTTING ASKFOR

General planning parameters Maximum total cost allowed Maximum tool suppliers allowed Maximum control suppliers allowed Number of unique machine suppliers Number of unique controller suppliers Overall configuration cost Processing machine Machine tool type Universal machine tool Lathe Milling machine tool Drdl Machine tool supplier Machine price (1065) Controller supplier

Table size (x-direction) Table size (y-direction) Table size (z-direction) Number of axes Processing accuracy Machine tool power (HP) Metal-cutting process Processing type Turning Milling Drilling Threading (internal) Threading (external) Threading (eccentric) Cutting Product to be processed Workpiece size (x-direction) Workpiece size (y-direction) Workpiece size (z-direction) Degrees of freedom required Processing accuracy required Metal-cutting operation needed Operation type Turning Milling Drilling Threading (internal) Threading(external) Threading (eccentric) Cutting Operation power required

R. Karni, A. Gal-Tzur

76

Table 3. A knowledge base for cell configuration (cont.) (c) Knowledge base: instance frames *INSTANCES FRAME FRAME SLOT SLOT SLOT SLOT SLOT SLOT SLOT SLOT SLOT SLOT FRAME SLOT

MAXIMUM CENTER- 1 MT-TYPE SUPPLIER PRICE CONTROL TABLE-X TABLE-Y TABLE-Z AXES ACCURACY POWER CENT- 1/ 1 PRO-TYPE

FRAME SLOT

CENT- 1/2 PROC

FRAME SLOT

CENT- 1/3 PRO-TYPE

FRAME SLOT

CENT- 1/4 PRO-TYPE

FRAME SLOT SLOT SLOT SLOT SLOT SLOT SLOT SLOT SLOT SLOT FRAME SLOT

MILL-1 MT-TYPE SUPPLIER PRICE CONTROL TABLE-X TABLE-Y TABLE-Z AXES ACCURACY POWER MILL- 1/ 1 PRO-TYPE

FRAME SLOT

MILL- 1/2 PRO-TYPE

FRAME SLOT SLOT SLOT SLOT SLOT SLOT SLOT SLOT SLOT SLOT FRAME SLOT

LATHE- 1 MT-TYPE SUPPLIER PRICE CONTROL TABLE-X TABLE-Y TABLE-Z AXES ACCURACY POWER LATH- 1/ I PRO-TYPE

FRAME SLOT

LATH- 1/2 PRO-TYPE

FRAME SLOT

LATH- 1/3 PRO-TYPE

FRAME SLOT SLOT SLOT SLOT SLOT SLOT SLOT SLOT

LATHE-2 MT-TYPE SUPPLIER PRICE CONTROL TABLE-X TABLE-Y TABLE-Z AXES

GLOBAL MACHINE PRESET PRESET PRESET PRESET PRESET PRESET PRESET PRESET PRESET PRESET PROCESS PRESET PARENT PROCESS PRESET PARENT PROCESS LITERAL PARENT PROCESS LITERAL PARENT MACHINE PRESET PRESET PRESET PRESET PRESET PRESET PRESET PRESET PRESET PRESET PROCESS PRESET PARENT PROCESS PRESET PARENT MACHINE PRESET PRESET PRESET PRESET PRESET PRESET PRESET PRESET PRESET PRESET PROCESS PRESET PARENT PROCESS PRESET PARENT PROCESS LITERAL PARENT MACHINE PRESET PRESET PRESET PRESET PRESET PRESET PRESET PRESET

CENTER MILICRON 0.4 FANUC 2.0 1.5 1.0 6 0.01 22

Upper constraints on solution Machining center 1 Universal machine tool

Process available on machine MILLING CENTER- 1 Process available on machine DRILLING CENTER- 1 Process available on machine INTHREAD CENTER- 1 Process available on machine OUTHREAD CENTER- 1 MILL MILICRON 0.1 FANUC 1.2 0 8 0.6 4 0.05 12

Milling machine 1 Milling machine tool

Process avadable on machine MILLING MILL- 1 Process available on machine DRILLING MILL-I LATHE MILICRON 0.10 FANUC 1.0 0 0.3 2 0.04 12

Lathe 1 Lathe

Process available on machine TURNING LATHE- 1 Process available on machine CUTTING LATHE- 1 Process available on machine EXTHREAD LATHE- 1 LATHE INGERSOL 0.15 IBH 1.5 0 0.7 2

Lathe 2 Lathe

Frame-based architectures for manufacturing planning an4 control Table 3. A knowledge base for cell configuration (cont.) SLOT SLOT FRAME SLOT

ACCURACY POWER LATH-2/1 PRO-TYPE

FRAME SLOT

LATH-2/2 PRO-TYPE

FRAME SLOT

LATH-2/3 PRO-TYPE

FRAME SLOT SLOT SLOT SLOT SLOT SLOT SLOT SLOT SLOT SLOT FRAME SLOT

DRILL- 1 MT-TYPE SUPPLIER PRICE CONTROL TABLE-X TABLE-Y TABLE-Z AXES ACCURACY POWER DRIL- 1/ 1 PRO-TYPE

FRAME SLOT

DRIL- 1/2 PRO-TYPE

FRAME SLOT SLOT SLOT SLOT SLOT FRAME SLOT SLOT

ITEM-1 SHAPE-X SHAPE-Y SHAPE-Z FREEDOM EXACT ITEM- 1/ 1 OP-TYPE ENERGY

FRAME SLOT SLOT

ITEM- 1/2 OP-TYPE ENERGY

FRAME SLOT SLOT SLOT SLOT SLOT FRAME SLOT SLOT

ITEM-2 SHAPE-X SHAPE-Y SHAPE-Z FREEDOM EXACT ITEM-2/1 OP-TYPE ENERGY

FRAME SLOT SLOT

ITEM-2/2 OP-TYPE ENERGY

PRESET PRESET PROCESS PRESET PARENT PROCESS PRESET PARENT PROCESS LITERAL PARENT MACHINE PRESET PRESET PRESET PRESET PRESET PRESET PRESET PRESET PRESET PRESET PROCESS PRESET PARENT PROCESS PRESET PARENT PRODUCT PRESET PRESET PRESET PRESET PRESET OPERATE PRESET PRESET PARENT PROCESS PRESET PRESET PARENT PRODUCT PRESET PRESET PRESET PRESET PRESET PROCESS PRESET PRESET PARENT PROCESS PRESET PRESET PARENT

0.05 10

INTHREAD 6 ITEM-2

EQ EQ EQ EQ EQ EQ EQ

REQUIRE REQUIRE RESOURCE RESOURCE RESOURCE RESOURCE RESOURCE

Process available on machine TURNING LATHE-2 Process available on machine CUTTING LATHE-2 Process available on machine EXTHREAD LATHE-2 DRILL MILICRON 0.05 FANUC 1.0 0.6 0.6 3 0.03 10

Process available on machine DRILLING DRILL- 1 Process available on machine INTHREAD DRILL- 1 Workpiece to be processed 0.6 0.5 0.3 3 0.07 Metal-cutting operation needed TURNING 9 ITEM- 1 Metal-cutting operation needed CUTTING 8 ITEM-2 Workpiece to be processed 0.9 0.4 0.4 4 0.05 Metal-cutting operation needed MILLING 11 ITEM-2 Metal-cutting operation needed

(d) Knowledgebase: rules *RULES

,

* Items and machines * in planning process SET LET LET LET LET LET LET LET * Conditions for * acceptable candidates

SCENARIO ITEM- 1 ITEM-2 CENTER-1 MILL- 1 LATHE- 1 LATHE-2 DRILL- 1

Drill 1 Drill

77

R. Karni, A. Gal-Tzur

78

SET IF AND AND AND AND AND THEN

ACCEPT PRO-TYPE MT-TYPE ACCURACY POWER TABLE-X TABLE-Z INCLUDE

IF AND AND AND AND AND AND THEN

PRO-TYPE ACCURACY POWER TABLE-X TABLE-Y TABLE-Z AXES INCLUDE

EQ EQ GE GE GE GE

OP-TYPE LATHE EXACT ENERGY SHAPE-X SHAPE-Z

EQ GE GE GE GE GE GE

OP-TYPE EXACT ENERGY SHAPE-X SHAPE-Y SHAPE-Z FREEDOM

* Conditions for * pruning configurations SET PRUNE LET VENDORS LET CONTROLS LET COST IF VENDORS OR CONTROLS OR COST

CU CU SM GT GT GT

SUPPLIER CONTROL PRICE MAXSUPP MAXCONT MAXCOST THEN EXCLUDE

* Conditions for * ranking configurations SET PREFER LET VENDORS LET CONTROLS LET COST RANK VENDORS RANK CONTROLS RANK COST

CU CU SM GT GT LT

SUPPLIER CONTROL PRICE

* END

Table 4. A knowledge base for manufacturing planning and control

(a) Paradigmdescription (1) Planning environment The environment is described by a set of interactive entities and their associated characteristics. These characteristics are in the form of connectances, or signals, which are passed between entities, thereby activating dynamic relationships between them. Connectances are represented by input/output slots associated with each entity, and the plan is initially in equilibrium. Scenario requirements are indicated by assigning deviation values to one or more connectances, which are then fixed, or "clamped' 31, for that scenario. The plan reacts to these deviations in such a way that further connectances may acquire devtation values m order to reach a new equilibrium. The aim of the system is to provide a qualitative simulation of plan behaviour in order to evaluate the appropriateness of the response and, if necessary, to modify the interrelationships between entities. (2) Entity representation (fact base) Each mteractwe entity is single-instance frame. Input and output connectances are represented by ternary slots with three possible values: '0' (zero, initial or planned range); '-' (negative, deviation below the planned range), and ' + ' (positive, deviation above the planned range). Zero thereby represents the planned or desirable state of the system; non-zero values indicate deviations from this state. (3) Problem-solving domain operators (rule base) Domain operators have the form of situation-action I F . . . THEN rules, partitioned into sets attached to a specific entity ('SET'). Each rule relates the values of an output slot to the values of one or more input slots, characterizing the behaviour of the entity in modifying output values as a function of Input values.

Frame-based architectures for manufacturing planning and control

79

(4) User interface (inference engine) The planner is requested to indicate which connectances are to be clamped and assigned deviation values and which are to be clam_ped. Potentially, all connectances can be modified or clamped; those selected are dictated by the nature of the scenario. (5) Control operators inference engine) (i) General sequence The inference mechanism requests the planner to indicate connectances to be modified and/or clamped. It then carries out multiple passes over all rule sets, in the order in which the frames are specified in the knowledge base ('INSTANCES'), modifying output values until a new equilibrium is reached. (ii) Rule set sequence The inference mechanism carries out multiple iterations over each set of rules for each frame instance referenced. At each iteration, rules are tested from top-to-bottom in the order in which they are defined in the knowledge base. As each rule is tested, it is skipped, confirmed or rejected. Rules which could change the values of clamped connectances are skipped. Rules which have been confirmed are ignored during subsequent iterations. Whenever a rule is confirmed and its consequent section activated, the iteration is terminated and testing for the subsequent iteration begins with the first remaining rule in the rule set. Iteration continues until no rule can be confirmed. (iii) Termination Termination occurs when connectance values cannot be modified further.

Table 4. A knowledge base for manufacturing planning and control (cont.)

(b) Knowledgebase: classframes * CLASSES FRAME SLOT SLOT SLOT

BUSINESS C-POLICY S-POLICY P-POLICY

TERNARY TERNARY TERNARY

OUTPUT OUTPUT OUTPUT

Business planning and control Capacity expansion policy Safety stock level policy Purchase order expedite policy

FRAME SLOT SLOT SLOT SLOT

CUSTOMER CO-DATES CO-ITEMS CO-DATES CO-ITEMS

TERNARY TERNARY TERNARY TERNARY

INPUT INPUT OUTPUT OUTPUT

Customer (orders and amounts) Sales plan (promised dates) Customer order quantity Sales plan (promised dates) Customer order quantity

FRAME SLOT SLOT SLOT

MARKET CO-DATES MS-DATES CO-DATES

TERNARY TERNARY TERNARY

INPUT INPUT OUTPUT

Marketing planning and control Sales plan (promised dates) Master schedule (delivery dates) Sales plan (promised dates)

FRAME SLOT SLOT SLOT SLOT SLOT SLOT SLOT SLOT SLOT SLOT

PRODUCE C-POLICY S-POLICY MS-LOAD CP-LOAD SF-LOAD SF-DATES MS-LOAD CP-LOAD SF-LOAD SF-LEAD

TERNARY TERNARY TERNARY TERNARY TERNARY TERNARY TERNARY TERNARY TERNARY TERNARY

INPUT INPUT INPUT INPUT INPUT INPUT OUTPUT OUTPUT OUTPUT OUTPUT

Production planning and control Capacity expansion policy Safety stock level policy Potential/maximum capacity Required/planned capacity Available/actual capacity Production schedule (actual dates) Potential/maximum capacity Required/planned capacity Available/actual capacity Shop order/production lead time

FRAME SLOT SLOT SLOT SLOT SLOT SLOT SLOT

MSPLAN CO-DATES CO-ITEMS MS-LOAD MR-DATES MS-ITEMS MS-DATES CP-LOAD

TERNARY TERNARY TERNARY TERNARY TERNARY TERNARY TERNARY

INPUT INPUT INPUT INPUT OUTPUT OUTPUT OUTPUT

Sales plan (promised dates) Customer order quantity Potential/maximum capacity Requirements plan (planned dates) Scheduled order quantity Master schedule (delivery dates) Required/planned capacity

FRAME SLOT SLOT SLOT SLOT SLOT

MRPLAN S-POLICY MS-LOAD MS-ITEMS MR-ITEMS MR-DATES

TERNARY TERNARY TERNARY TERNARY TERNARY

OUTPUT INPUT INPUT INPUT INPUT

Material requirements planning Safety stock level policy Potential/maximum capacity Scheduled order quantity Planned order quantity Requirements plan (planned dates)

R. Karni, A. Gal-Tzur

80

Table 4. A knowledge base for manufacturing planning and control (cont.) SLOT SLOT SLOT SLOT SLOT SLOT SLOT SLOT SLOT SLOT SLOT

MS-DATES PO-LEAD SF-LEAD PO-END SF-END SF-ITEMS PO-ITEMS SAFETY MR-DATES MR-ITEMS SF-LEAD

TERNARY TERNARY TERNARY TERNARY TERNARY TERNARY TERNARY TERNARY TERNARY TERNARY TERNARY

INPUT INPUT INPUT INPUT INPUT INPUT INPUT INPUT OUTPUT OUTPUT OUTPUT

Master schedule (delivery dates) Purchase order/supplier lead time Shop order/production lead time Purchase order arrival Shop order completion Production output quantity Supplier delivered quantity Safety stock level Requirements plan (planned dates) Planned order quantity Shop order/production lead time

FRAME SLOT SLOT SLOT SLOT SLOT

CRPLAN MS-LOAD CP-LOAD SF-LOAD SF-LEAD SF-DATES

TERNARY TERNARY TERNARY TERNARY TERNARY

INPUT INPUT INPUT INPUT OUTPUT

Capacity planning and scheduling Potential/maximum capacity Required/planned capacity Available/actual capacity Shop order/production lead time Production schedule (actual dates)

FRAME SLOT SLOT SLOT SLOT SLOT

SHOPFLOR SF-DATES SF-LEAD SF-END SF-ITEMS SF-LOAD

TERNARY TERNARY TERNARY TERNARY TERNARY

INPUT INPUT OUTPUT OUTPUT OUTPUT

Shop floor operation and control Production schedule (actual dates) Shop order/production lead time Shop order completion Production output quantity Available/actual capacity

FRAME SLOT SLOT SLOT SLOT SLOT SLOT SLOT SLOT SLOT

PURCHASE P-POLICY MR-DATES MR-ITEMS SF-ITEMS PO-LEAD PO-END PO-ITEMS PO-END PO-ITEMS

TERNARY TERNARY TERNARY TERNARY TERNARY TERNARY TERNARY TERNARY TERNARY

OUTPUT INPUT INPUT INPUT INPUT INPUT INPUT OUTPUT OUTPUT

Purchasing operation and control Purchase order expedite policy Requirements plan (planned dates) Planned order quantity Production output quantity Purchase order/supplier lead time Purchase order arrival Supplier delivered quantity Purchase order arrival Supplier delivered quantity

FRAME SLOT SLOT SLOT SLOT ,

INVENTRY SF-ITEMS PO-ITEMS CO-ITEMS SAFETY

TERNARY TERNARY TERNARY TERNARY

INPUT INPUT OUTPUT OUTPUT

Inventory operation and control Production output quantity Supplier delivered quantity Customer order quantity Safety stock level

FRAME SLOT SLOT SLOT

SUPPLIER PO-ITEMS PO-LEAD PO-END

TERNARY TERNARY TERNARY

INPUT OUTPUT OUTPUT

Supplier (orders and amounts) Supplier delivered quantity Purchase order/supplier lead time Purchase order arrival

(c) Knowledge base: instance frames *INSTANCES FRAME FRAME FRAME FRAME FRAME FRAME FRAME FRAME FRAME FRAME FRAME

COMPANY DEMAND SALES PLANNING MASTER MATERIAL CAPACITY WORKSHOP ACQUIRE STORES VENDOR

BUSINESS CUSTOMER MARKET PRODUCE MSPLAN MRPLAN CRPLAN SHOPFLOR PURCHASE INVENTRY SUPPLIER

Business planning and control Customer (orders and amounts) Marketing planning and control Production planning and control Master schedule planning Material requirements planning Capacity planning and scheduling Shop floor operation and control Purchasing operation and control Inventory operation and control Supplier (orders and amounts)

Frame-based architectures for manufacturing planning and control

81

(d) Knowledgebase:ru~s *RULES

SET IF THEN IF THEN

MARKET MS-DATES CO-DATES MS-DATES CO-DATES

SET

PRODUCE

IF AND OR OR THEN

C-POLICY CP-LOAD SF-LOAD SF-DATES MS-LOAD SF-LOAD CP-LOAD SF-LOAD

EQ + EQEQ EQ + EQ + EQ 0 EQ 0 EQ 0

IF AND AND OR OR THEN

C-POLICY MS-LOAD CP-LOAD SF-LOAD SF-DATES SF-LEAD

EQ NE EQ EQ EQ EQ

SET

MSPLAN

IF THEN IF THEN IF THEN

CO-ITEMS MS-ITEMS CO-ITEMS MS-ITEMS CO-ITEMS MS-ITEMS

EQ + EQ + EQEQ EQ 0 EQ 0

IF THEN

MS-LOAD CP-LOAD MS-DATES

EQ EQEQ +

IF THEN IF THEN

MR-DATES MS-DATES MR-DATES MS-DATES

EQ EQ NE EQ

+ + + 0

SET

MRPLAN

IF THEN IF AND THEN

MR-ITEMS MR-ITEMS MR-DATES MS-LOAD MR-DATES

EQ EQ NE NE EQ

0 MS-ITEMS + MS-DATES

IF THEN IF AND THEN

SF-ITEMS MR-ITEMS SAFETY S-POLICY MR-ITEMS

EQ EQ + EQEQ + EQ +

IF AND THEN

MR-ITEMS MS-LOAD SF-LEAD

EQ + NE + EQ +

EQ EQ NE EQ

+ + + 0

0 + + +

IF AND THEN IF THEN IF OR AND AND THEN IF AND THEN * SET IF THEN * IF OR THEN * IF AND THEN IF AND AND AND THEN * SET * IF THEN IF AND THEN * IF THEN * SET * IF AND THEN * IF AND OR THEN * IF AND THEN * SET * IF THEN * *END

SF-LEAD MS-LOAD MR-DATES PO-LEAD MR-DATES SF-LEAD PO-LEAD SF-LEAD PO-LEAD MR-DATES SF-LEAD PO-LEAD MR-DATES

EQ + NE + EQ + EQ + EQ + EQEQNE + NE + EQEQ 0 EQ 0 EQ 0

CRPLAN SF-LEAD SF-DATES

EQ + EQ +

MS-LOAD CP-LOAD SF-DATES

EQEQEQ +

MS-LOAD SF-LOAD SF-DATES CP-LOAD SF-LOAD MS-LOAD SF-LEAD SF-DATES

EQEQEQ + NENENENE + EQ 0

SHOPFLOR SF-DATES SF-END SF-DATES SF-LEAD SF-END

EQ EQ EQ NE EQ

SF-LEAD SF-END

EQ + EQ +

+ + 0 + 0

PURCHASE MR-ITEMS SF-ITEMS PO-ITEMS

EQ + EQ 0 EQ +

P-POLICY PO-LEAD PO-END PO-END

EQ EQ EQ EQ

P-POLICY PO-LEAD PO-END

NE + EQ + EQ +

+ + + 0

SUPPLIER PO-ITEMS PO-LEAD

NE 0 EQ +

82

R. Karni, A. Gal-Tzur

2 PLANNING SYSTEM ARCHITECTURE

(4) Control operators (inference engine)

Zozaya-Gorostiza et al.4° present architectural requirements for knowledge-based process planning systems. They classify these factors with respect to four perspectives:

The inference mechanism determines which potential action the AI system should perform at each point in plan development. It thus determines the order in which domain operators are referenced, and initiates user communication and domain operator execution.

(a) Knowledge representation: how planning knowledge is stored (b) Problem-solving operators: how planning knowledge is used to perform planning tasks (c) User interaction: how users can direct the inference process, modify knowledge or override decisions (d) Control operators: how the problem-solving operators are to be applied to obtain a plan We utilize this approach to define four aspects of the frame-based planning systems described in this article: (1)

Entity representation (fact base)

Information regarding objectives or constituents of the planning process -- factors impacting the plan (goals, constraints, environmental factors) and the plan itself (activities and resources) -- is represented in frames or templates. Each object or frame has a unique name defining it. It is first defined as a prototype or class frame. Then, actual instances of frames, which are individuated copies 35 of each frame class to be used solving the planning problem, are defined. There may be only one instance of a unique frame class (as, for example, a specific instance of an environmental impact), or there may be several instances of a given frame class (such as various stages in a plan). Each frame contains slots which store values of the object attributes. Slots may also hold references to other frames, thus enabling linkages between frames and inheritance of slot values from these frames. Slots are defined by a unique name, a structure (literal, binary, ternary, numeric or frame), and the mechanism by which the slot value will be attained (by direct communication with the user or via a procedural attachment). Slot values are usually initialized to 'NULL' or zero or given some default value ('PRESET'). A 'GLOBAL' frame acts as a universal parent, representing the planning scenario and other general planning information accessible to all other frames. (2) Problem-solving domain operators (rule base) Rule-based functions or procedures, usually in modular or partitioned form, perform the planning tasks by creating or modifying slot values. They explicate the relations between frames, and are usually domainspecific. (3) User interface (rule base and inference engine) Communication with the planner enables him/her to input situation-specific data or adapt or modify system knowledge.

3 AN EXPERT SYSTEM FOR DESIGNING AN OPERATIONS REGIME

Operations management is concerned with integrating and coordinating those systems concerned with producing and delivering the products provided by an organization: 21"26'27(1) the product (design); (2) marketing; (3) process (technology); (4) extraction (raw materials); (5) transportation (raw material and finished product delivery); and (6) maintenance (product performance feedback). In particular, the operations function is responsible for the smooth and timely flow of materials from raw materials from suppliers through finished goods handed over to the customer; for implementing efficient and effective time-phased allocation of resources; for scheduling jobs; and for managing inventories. Thus the manner in which operations are planned and executed is crucial for success. Operations regime design is the procedure by which the physical, operational and organizational frameworks of production are specified such that company goals and constraints are satisfied. 26'27 This requires making decisions as to the strategies to adopt regarding the attributes of such a regime. For example: (1)

Material flow

(2) Added capacity (3) Inventory levels (4) Safety stock (5) Work scheduling (6) Lot sizing (7) Logistics

jobshop? flowshop? group technology? flowline? overtime? shifts? sub-contracting? make-to-stock? make-up-to-stock? make-toorder? none? some? much? OP/OQ? MRP? JIT? OPT? fixed orders? lot-for-lot? optimized? company vehicles? customer/supplier vehicles?

These decisions are contingent upon the nature of the other systems. For example: (1) (2) (3)

Product Product cycle Market

(4) Dehvery (5) Raw materials (6) Supplier,, (7) Technology

standard or special? simple or complex 9 new hem? regular item? obsolescent item? stable or changing? favourable or competilwe '~ localized or dispersed'? regular or irregular? standard or special? simple or complex? many or few 9 flexible or rigid? stmple or complex 9 regular or innovative?

Naturally, each system has a large number of attributes and alternatives, and also a large number of functional characteristics 9~°. Enumerating all the possible interrelationships presents an enormous problem. However, a frame-based system enables a start to be made in developing an operations regime design tool which takes a wide range of factors into account.

Frame-based architectures for manufacturing planning and control In this paper we present a simplified illustration of an operations regime design procedure, concentrating on four related systems: (1) PRODUCTS (2) MARKET (3) SUPPLIES (4) PROCESS

Product/mix characteristics Product market characteristics Suppher/raw material characteristics Production process characteristics

An operations regime attributes: (I) (2) (3) (4) (5) (6) (7)

CELLTYPE SHIFTS FINISHED SCHEDULE LOTSIZE SAFETY DELIVERY

is characterized by

seven

Production cell organizatton Necessity for shifts/overtime Product inventory control Scheduling methodology Size of production lots Safety stock required Transportation of products

Rules have been constructed by taking one of the options (usually the most conservative or complex) as default, and then specifying circumstances in which the default option would not be advisable. When consulting the system, the designer provides information regarding the specific attributes of the four environmental systems. The rules generate recommended characteristics of the operations regime.

4 AN EXPERT SYSTEM FOR PROJECT PLANNING Because of the non-repetitive nature of projects, which therefore require special allocations of time and resources, successful project planning requires detailed knowledge and extensive experience regarding the tasks to be carried out, deadlines to be met, and the conditions under which tasks are to be performed. 3"4"5"6'15"22"28"29" 30.3.s.36 These conditions influence expected durations, resource commitments and scheduling priorities. The final project program, usually represented as a project network, is thus the outcome of an expert process by which the project manager creates a plan which will meet as many goals and constraints as possible. Conventional project management tools require a complete specification of the project network, with finalized time and resource allocations, as input. They do not, however, enable the planning knowledge which has been used to create the network to be explicitly represented and retained. An intelligent project management system (IPMS) allows the project planner to develop a plan more in accordance with the way planning is usually carried out. He/she stipulates activities to be performed, their work content, resources available and restrictions on their use, policies for setting priorities, milestones to be achieved, deadlines to be met, and so on. The IPMS then generates a project network or plan which most closely meets all these situations. To do this, it may incorporate aspects and relationships between them such as the following (compare Refs 3,4,5,6,15,22,27,28,29,35,36,38):

83

(1) (2)

Processes: activities and milestones Product components: outputs or results of activities -- documents, specifications, deliverables (3) Resources: used to perform activities -- manpower, equipment (4) Constraints and goals: deadlines, availabilities, environmental characteristics (5) Temporal implications: technological precedences, overlaps (6) Causal implications: logical precedences, conflict resolution policies, priory setting These entities can be represented by frames, and the relationships between them by rules. The power and versatility of such a system is illustrated by the development of a planning system for the launching of a new product. ~ In modern manufacturing, with small lots and rapid obsolescence of products, many production plans are developed using project management, rather than repetitive manufacturing, techniques. The planner will have a set of typical activities related to launching a new product, ~ and will particularize them and their time and resource requirements to a specific project. The resultant project network generation and solution is developed as follows: (I) The project 'scenario' (frame: SCENARIO; rule: SETTING) The project scenario defines the setting in which the project is to be carried out. It details specific characteristics and conditions which are likely to influence goals and allocations. In our example the planner defines the level of technological complexity of the product to be developed and the process to manufacture it, market competitiveness and the time frame for launching the product. These impact the ease with which product and process design can be carried out. (2) Time windows WORKSHOP)

(frames:

VACATION,

Time windows, defined by a start and end date, constitute constraints on the availabilities of resources and thus the possibility of performing certain activities. In our example, we define two time windows -- vacation time and training time -- when certain teams may not be available to the project.

(3) Causalities (frames: TREE, MAKE) Causalities are constraints or circumstances which affect resource or time allocations to activities. For example, time add-ons, shown as a multiplicative factor, reflect the level of product and production complexity.

(4) Resources (frames: MARKET, DESIGN, PROCESS) Resources are used to perform activities. Characteristics of each resource may include a default and maximum

84

R. Karni, A. Gal-Tzur

intensity of use, time add-on conditions, and time intervals or windows during which the resource is unavailable.

(5) Milestones (frame: FEASIBLE) Milestones indicate 'must meet' or 'should meet' dates for completing certain activities or groups of activities. For example, the milestone may represent a decision point for continuing with product development -- or abandoning the project -- when one of the project phases has been completed.

(10) Insertion of static logical predecessors (rule: LOGIC) One important advantage of IPMS is that the planner is not required to make all precedences explicit. Implicit precedences may be static or dynamic. Static precedences may be explicated and added to the precedence list of each activity. In our case, we look for 'output/input' precedences; if an activity requires a certain product component as input, it must succeed the activity which produces the component as an output.

(11) Insertion of dynamic logical predecessors (rules: (6) Activities (frames: P-DESIGN, P-SALES, METHODS) Activities are steps by which the project is carried out. In addition to the conventional precedences, duration, resource type and levels, and planned or computed start and end dates, and IPMS environment adds in factors which may influence sequencing, duration and resource requirements, and windows available for executing the activity, such as: project phases and associated milestones; work content and resource intensity; product components as inputs to and outputs from activities; conditions under which the resource allocation may be modified or be unavailable; resource allocation conflict priorities; and the ability to add further 'logical' precedences. Three typical activities are illustrated in the knowledge base.

(7) Initial resource and time allocation (rule: ALLOCATE) The planner indicates those parameters (work content and/or duration and/or intensity) which are most normal for him, and the IPMS computes the remaining parameters using the relationship (work content) = (duration) x (intensity). Initially, the given or default intensity is used. If the activity is to be 'crashed', the system increases the intensity level and reduces the duration accordingly.

(8) Priority setting (rule: PREFER) The planner establishes policy rules to set priorities for activities competing for resources. In our example, we demonstrate two policies: activities requiring high resource levels are given the highest priority; activities associated with the initial study phase (preceding a project feasibility decision) are given intermediate priority. (9) Scenario-related resource and time allocation (rule: ADJUST) The planner establishes circumstantial rules for readjusting resource and time allocations, in accordance with the project scenario or other policy factors. In our example: if the market is competitive, then manpower is to be added to marketing-related activities; durations of product-design-related activities are to be multiplied by a factor related to product complexity; and durations of production-design-related activities are to be multiplied by a factor related to process complexity.

C O N F L I C T , SHIFT) In our example, we show the implicit insertion (by time shifting) of two dynamic precedences i.e. precedences which are only activated if the activity sequence contradicts a constraint. If two activities compete for a resource, that with lower priority is delayed until the other has been completed; if activity dates fall within a disabling time window, the activity is delayed until the end of the time window.

(12) Dynamic resource and time reallocation (rules: T A R G E T - l , TARGET-2) If the computed activity dates do not meet deadlines requirements, the I P M S , in accordance with 'crashing' policy, automatically tries to readjust resources and time in order to fulfil these constraints. If the activity is "crashable', then the time window which resources are normally unavailable is cancelled. If this is not sufficient to meet the deadline, then manpower, is available, is added to shorten the activity duration.

(13) Scheduling calculations (rules: RESET, F O R W A R D , C O N V E R G E , CHANGED) Schedules are calculated using the conventional right shift algorithm. If resource or time window conflicts occur, their resolution results in further right shifting. This is repeated until the solution converges. Then, if deadline conflicts occur, resolving them results in reduced activity durations -- or implied left shifts. This requires schedule dates to be re-initialized for all activities, and the total scheduling pass repeated, until no further durations are reduced or conflicts occur. 5 AN E X P E R T SYSTEM F O R CONFIGURATING A PRODUCTION CELL Modern manufacturing is often organized into a group technology pattern, where groups of machines, or cells, provide a compendium of processes for a given product mix. A cell may be composed of assemblage of existing machines -- from within a larger group of machines -dedicated to processing the products, and possibly linked by a computer-controlled network so as to constitute and operate the cell. Alternatively, the designer may be required to specify a new set of machines to be expressly purchased and linked for this purpose. Obviously, the proposed cell must be capable of carry-

85

Frame-based architectures for manufacturing planning and control

ing out all operations required for the specified product mix. Moreover, the types and sources of the machines comprising the cell should be favourable to cell management, especially in a computer-integrated manufacturing environment, so that tooling, setup, operation, monitoring and maintenance of the cell are facilitated. In order to develop a satisfactory configuration, the designer must take a large number of factors into consideration:

preferred.

(1)

(3)

(2)

(3)

(4)

(5)

All relevant properties of the products, such as: shapes, sizes, tolerances All relevant operations on the products, such as: operation type, tool type, degrees of freedom required, power needed All relevant properties of candidate machines, such as: machine class, power, axes, accuracy, processes that can be performed, tools that can be handled Criteria for evaluating alternative 'competing configurations', such as: cost, number of different machine tool suppliers, number of different machine controller supplies Limitations on acceptable configurations, in terms of budgets, controllability and maintainability

The basic configuration problem, then, is to choose the best cell, out of a given candidate set of existing machines or machines to be purchased, which is capable of processing a given set of products. The problem encompasses two entities -- machines and products -and several decision points regarding the acceptability and desirability of potential, or 'competing', cell configurations. These decision points are specified by the designer as sets of rules, which are used to develop alternative machine groupings and eventually select that most

(1) (2)

(4)

(5)

(6) (7)

First, the product mix and candidate machines is specified ('SCENARIO' rule set). The product mix implicitly generates a list of operations which are to be performed; likewise, the machine set generates a list of processes which can carry out these operations. Typically, any single machine can only carry out some of the operations required. An initial set of partial 'competing configurations' is established, such that each configuration is capable of carrying out the first operation in the list. This capability is specified and tested by the 'ACCEPT' rule set. The next operation in the list is checked against all 'competing configurations', using the same 'ACCEPT' rule set. If it cannot be carried out by some partial configuration, that machine group is augmented by a machine capable of executing the operation. As several alternative machines may be able to do so, several additional 'competing configurations' may be generated. Each partial configuration is then tested to see whether it exceeds limitations imposed on overall acceptability. It is eliminated if these constraints are violated ('PRUNE' rule set). The procedure is repeated until all operations have been covered. All complete 'competing configurations' are then ranked in descending order of preference ('PREFER' rule set).

Generating and selecting cell configurations for the illustrative knowledge base (Table 3) provide the following outcomes:

Step

Product

Operation

Group 1

Group 2

Group 3

Group 4

1 2 3

ITEM- 1

ITEM- 1/ 1 ITEM- 1/2 ITEM-2/1

LATHE- 1 LATHE- 1 LATHE- 1 CENTER- 1 LATHE- 1 CENTER- 1

LATHE-2 LATHE-2 LATHE- 1 MILL- 1 LATHE- 1 MILL- 1 DRILL- 1

LATHE-2 CENTER- 1 LATHE-2 MILL- 1 DRILL- 1

LATHE-2 MILL- 1

ITEM-2

4

ITEM-2/2

The following limitations are imposed: maximum cost: $500,000; maximum machine suppliers: 2; maximum controller suppliers: 2. At step 3 the fourth group is eliminated on cost. The final three 'competing configurations' have the following characteristics:

Group

Group 1 Group 2 Group 3

Machine

Controller

suppliers

suppliers

Total cost

1 1 2

1 1 2

$500,000 $300,000 $350,000

Group 2 (LATHE-I, MILL-I, DRILL-I) would be the preferred design. 6 AN EXPERT SYSTEM FOR ANALYSIS OF MANUFACTURING OPERATIONS

A modern manufacturing framework is composed of planning and execution which work together to create and execute a feasible, and, hopefully, efficient, production plan. Such a manufacturing planning and control (MPC) environment is discussed by Vollman, et al. 39 They visualize three groups of components, or subsystems, those for policymaking and direction setting,

86

R. Karni, A. Gal-Tzur

those for planning requirements to achieve production goals, and those for executing these plans. We can utilize this breakdown to provide a schematic of a typical MPC system encompassing the following components: (1) Policymaking and direction setting o Business planning o The market of customer o Marketing or demand planning o Production and resource planning (2) Requirements planning o Master production scheduling and aggregate capacity planning o Materials requirements planning o Capacity requirements planning and scheduling (3) Executing o Shop floor control o Inventory control o Purchasing o Suppliers and vendors In order to maintain a smooth production schedule in the face of problems and changes, a large amount of interaction has to take place between these sub-systems. It is well known that the lack of a holistic approach to production planning is one of the main causes of poor performance (compare Refs 1,2,8,9,16,18,19,34). However, it is almost impossible to visualize an overall model which will not only incorporate the detailed functioning of each of the sub-systems and the knowledge passed between them, but also enable the planner to study the effects of changes on their behaviour. Even 2° lists the inputs to, transformations by and outputs of several sub-systems, and draws up a map connecting these sub-systems together. Burbidge 891° goes further and describes in detail a 'connectance' model for production variables. He classifies over 200 variables involved in a production system. In order to be able to study interactions, while avoiding the complexities of formulating detailed quantitative relationships, he proposes a qualitative approach which deals only with 'directions of change': how a given negative or positive change in one or more variables induces a particular. direction of change in another variable (positive or negative connectances). Eloranta et al.~8"~9 outline a combined quantitative (simple arithmetic transformations) and qualitative model, composed of entities (facilities, departments, cells, items), interactions (input, output), and descriptive variables (lead times, costs, revenues). The detailed structure and operation of a "qualitative reasoning' model of this type -- and expert system for fault diagnosis -- is described by Shaw & Menon. 37 The qualitative reasoning approach avers that a suitably constructed intelligent system can constitute a useful approximation to the behaviour of a more complex system, using relatively simple computations and reasoning. In particular, it characterizes each system element by how it maps input into output variable values, and enables 'paths of causal interaction' between

system variables to be studied. Variables have a ternary structure, obtained by mapping quantitative values into qualitative equivalents across the entire value space for the variable. These qualitative values are: zero (0) for values within the desired or planning range; negative ( - ) for values below this range; and positive (+) for values above this range. A reasoning process, 'envisionment', enables the overall system behaviour to be predicted by propagating a set of positive and/or negative perturbations through the interaction paths, using rules for each system element to transform variable values. In this way, what-if questions about the effects of the state of a component on overall system behaviour can be answered. A similar qualitative reasoning model is incorporated into a knowledge-based process control system described by Egilmez & Kim. ~7 Eloranta, et al. t9 refers to models of this type as 'qualitative simulation' models. We have used this approach to formulate an intelligent model of a manufacturing planning and control system. Some twenty variables involved in policymaking, planning, scheduling and executing have been related to several sub-systems, both as inputs to some sub-systems and outputs from others. In this way, interactions and the manner in which the MPC system reacts to external and internal changes can be studied. The rules, which in this model constitute an approximation of the workings of eleven MPC entities, indicate how they act or react to changes in input variables to produce changes in output variables: (1)

(2)

(3)

(4)

(5)

(6)

(7)

Business planning and control (BUSINESS) o Set policies (capacity expansion, safety stock, purchase expedite) Customer (SERVICE) o Set delivery dates o Set order quantities Marketing (MARKET) o Adapt delivery dates to planning/execution capabilities Production planning and control (PRODUCE) o Provide added capacity, if permitted o Increase production lead times if no added capacity allowable Master schedule planning (MSPLAN) oAdjust scheduled order quantities to customer demand o Delay delivery dates if maximum planned capacity insufficient oDelay delivery dates if production or supply delayed Material requirements planning (MRPLAN) o Coordinate lot sizes with the master schedule o Increase lot sizes if production is short or safety stocks are low o Increase lead times if lot sizes are increased oAdjust due dates according to change in lead times Capacity planning and scheduling (CRPLAN) o Delay scheduled dates if lead times increased

Frame-based architectures for manufacturing planning and control o Delay scheduled dates if maximum planned capacity insufficient o Delay scheduled dates if actual capacity insufficient (8) Shop floor operation and control (SHOPFLOR) o Adapt order release dates to schedules dates o Delay order completion if lead times increased (9) Purchasing operation and control (PURCHASE) oIncrease order quantities if requirements increased o Expedite to overcome lead time increases, if permitted o Delay supply dates if expediting not permitted Several scenarios indicate how the model can be used to study the effects of policies and disturbances in manufacturing planning and control: Insufficient capacity (with or without added capacity permitted) (2) Production shortfall (with or without added capacity permitted) (3) Delay in supply (with or without expediting permitted)

In a previous article. 25 the authors have detailed six further paradigms useful for planning tasks: (!)

(2)

(3)

(4)

(1)

The outcomes -- variables showing deviations at the end of a scenario run -- may be summarized as follows:

87

(5)

A logical expert system for material handling equipment: the manufacturing environment is described by yes/no attributes; the system selects a specific device A multi-valued expert system for automated guided vehicle control: the manufacturing environment is characterized by discrete multi-valued attributes; the system selects an appropriate dispatching rule A profile matching expert system for project management software: the project, software and hardware environments are expressed by discrete multi-level attributes; the system recommends that package closest to a desired profile A confidence-building expert system for machine feed equipment: the operating environment is defined by unsymmetric yes/no attributes; the system develops a level of confidence for the suitability of each device A tandem decision system for production scheduling: the production environment is defined by a mathematical programming model coupled with rules defining 'fuzzy' constraints and solution failure recovery; the system determines the proportions of parts to be allocated to alternative process rootings

Scenario

Disturbance variable

Policy variable

Deviation variables

1

SF-LOAD ( - )

C-POLICY (+) C-POLICY (0)

MS-LOAD SF-LEAD SF-END SF-DATES MR-DATES MS-DATES CO-DATES

(+) (+) (+) (+) (+) (+) (+)

2

SF-ITEMS ( - )

C-POLICY (+)

MR-ITEMS MS-LOAD MR-ITEMS SF-LEAD SF-DATES MR-DATES MS-DATES CO-DATES

(+) (+) (+) (+) (+) (+) (+) (+)

C-POLICY (0)

3

PO-LEAD (+)

7 RELATING MPC TASKS TO FRAME-BASED

ARCHITECTURES We have demonstrated that frame-based architectures provide a suitable basis for developing knowledge-based systems in manufacturing planning: frames represent planning entities; rules express interrelationships; and the planning strategy is mirrored in the inference process.

P-POLICY (+) P-POLICY (0)

(no deviations) PO-END MR-DATES MS-DATES CO-DATES

(+) (+) (+) (+)

(6) A situation-action expert system for controlling job-to-machine allocation in a flexible manufacturing cell: product, machine and dispatching rule properties are integer valued; the system allocates machines to jobs as they arrive at the cell These paradigms have a Simpler structure than those portrayed in this article: the knowledge base is composed of a single list of facts and single rule set, rather

88

R. Karni, A. Gal-Tzur

than being partitioned into frames and rule sets; and the inference process is determined entirely by the rule base. The aim of the first four examples is to find the best of a predetermined set of outcomes, based on a general set of attributes. Structure in the fifth example is provided by a linear programming matrix, which is manipulated both by the simplex algorithm (variables) and the rule base (constants). In the last example both jobs and machines are characterized by only two attributes: operation times (input) and completion dates (output). The representation is adequate for the planning problems described. In general, we need to be able to categorize these ten planning architectures. Such a scheme is provided by Chandrasekeran, ~2"~3 who classifies AI-based systems in terms of generic tasks. His goal is to identify 'building blocks' of reasoning strategies which are both general and widely useful. The advantages of this approach are: (a) it recognizes that multiformity of different problems and the necessity for variegated strategies; (b) it exploits and integrates the interrelationships between knowledge and inference; (c) it facilitates knowledge acquisition through a generic-task-related acquisition strategy; (d) it helps build generic-taskrelated explanation facilities; and (e) overall, it facilitates the design and implementation of knowledgebased systems. A generic task is characterized by: (a) functionality or goal of the task; (b) input-output behaviour; (c) representation and organization of knowledge; and (d) inference and control process. Six generic tasks have been identified: 12.13.24 hierarchical classification, hypothesis matching, object synthesis, abductive hypothesis assembly, state abstraction and knowledgedirected information passing. Characteristics of these six tasks are summarized in Table 5. (Types 1 to 6). The ten MPC systems described in this and the previous article 25 can be related to these generic categories as follows:

MPC System

Generic task

Equipment selection AGV dispatching rule selection PM software selection Machine feeder selection Production schedule development FMS job scheduhng

Hierarchical classification Hierarchical classification Hypothesis matching Hypothesis matching Object synthesiss

Production regime design Project planning Process configuration Manufacturing planning & control

Knowledge-directed transaction passing Knowledge-directed information passing Object synthesis Abductive hypothesis assembly State abstraction

We see that the paradigms can be usefully classified by Chandrasekaran's generic task scheme. It requires the addition of a seventh generic task, knowledgedirected transaction passing, to describe those systems

which are required to respond repetitively to timewise or staged inputs. This task is defined in Table 5 as type 7. We also see, as pointed out by Oxman, 32 that the inference mechanism is an integral part of the generic task 'package', rather than an entity apart from the knowledge representation of the task. The scheme is therefore an effective starting point for characterizing MPC systems and associating them with appropriate architectures, and demonstrates the generality of problem structures encountered in different fields. Regarding the number of possible generic tasks, Chandrasekaran ~2 states that 'the integrated generic task toolset is extensible in the sense that more generic tools can be added as they are invented, and additional problem solvers can be invoked as needed'. Johnson & Zualkernan, 23 in commenting on Chandrasekaran's scheme, assert that there indeed exists a profusion of knowledge-based system types covering a large number of applications, so that such a scheme could help control this problem. However, 'there does not yet appear to be a m e a n s . . , for deciding what organization or structure among a set of generic tasks is most suitable for a specific problem-solving (environment)'.

8 DISCUSSION We have presented four flame-based architectures to demonstrate how knowledge-based techniques can be effectively employed in the design, planning, synthesis and analysis of manufacturing systems. The main features of these paradigms are summarized in Table 6. They enable the problem to be decomposed into three parts: entities (environment, resources, activities, plans) - - represented by frames: relationships (interactions, functions)-- expressed by rules; and planning and design strategies -- activated by the inference engine. These strategies are classifiable in terms of generic tasks. Thus a suitable architecture may be selected for a given task by categorizing the problem (design, plan, synthesize, analyze); characterizing the plan structure (single or multiple components); relating the rules (to slots or frames or groups of frames); and selecting the appropriate inference mechanism (slot, rule or procedurally driven). Small-scale systems have been presented in the illustrative knowledge bases. Several problems may arise when full-scale knowledge bases are developed: (1) P r o d u c t i o n regime design

Potentially, an enormous number of rules are required to interrelate all feasible combinations of environmental attributes to the appropriate features of an operations regime. The approach used here was to select the most complex alternative as a default, and specify rules for recommending less severe options. A proper procedure for developing rules for this type of knowledge-directed information passing paradigm needs to be developed, to avoid falling into a 'combinatorial trap'.

Frame-based architectures f o r manufacturing planning and control

89

Table 5. Generic tasks

(1) Hierarchical classification Goal Input Output Knowledge Inference

Given a situation, determine what hypotheses apply to it Description of the situation Hypotheses applying to the situation Classification hierarchy: hypothesis types and sub-types, how well situation data matches types and sub-types Establish nodes in hierarchy; refine by descending hierarchy; backtrack on failure

(2) Hypothesis matching Goal Input Output Knowledge Inference

Given a situation, determine which hypothesis applies best to it Description of the situation How well hypotheses match the situation Evidence abstraction hierarchy: hypothesis types and sub-types; how well situation data matches types and sub-types Compute degree of confidence at each level and aggregate confidence

(3) ObJectsynthesis by selection and refinement Goal Input Output Knowledge Inference

Design an object satisfying given specifications Specification of object to be designed Design meeting the specification Hierarchy mirroring the structure and components of the object; precompiled partial (parametric) or total component alternatives; failure recovery mechanisms; how to select the most appropriate component Choose partial plans at each level; backtrack on failure

(4) Abductive hypothesis assembly (composite hypothesesformation) Goal Input Output Knowledge Inference

Find a set of hypotheses which together explain a given situation Description of the situation; hypotheses, each of which covers some aspect of the situation Composite hypothesis best covering the situation Relationships between hypotheses and situations; relative significance of the data describing the situation Assemble hypotheses progressively to explain most significant data; critique to remove superfluous hypotheses

(5) State abstraction Goal Input Output Knowledge Inference

Given a change in a system, indicate the consequences on the behaviour of the system The initial state of the system and the changes imposed Final situation of the system elements and connections Description of system element behaviour and connections between elements Follow through changes in states and relationships between system elements, via connections between them

(6) Knowledge-directedinformation passing Goal Input Output Knowledge Inference

Given some situation data, determine all relevant data Partial description of the situation Total description of the situation Hierarchy of data object types and sub-types; how remaining data can be derived (defaults, attachments) Use knowledge-intensive procedures to obtain all data

(7) Knowledge-directedtransaction processing Goal Input Output Knowledge Inference

Process a given set of transactions such that a defined objective is achieved The set of transactions, sequenced or unsequenced, and the initial state of the system The sequence in which the transactions are processed, and the process applied to each one Interrelationships between system states, processes and transactions; how to determine the appropriate processing sequence and process Select a sequencing and processing strategy at each stage; follow through changes in system states and transactions

90

R. Karni, A. Gal-Tzur Table 6. Planning and design tasks

Plan/design task

Task achvity

Generic task Frames

Production regime

Design

Project planning

Planning

Processing configuration Operations planning & control

Synthesis Analysis

Knowledge directed information passing Object synthesis

Composite hypothesis formation State abstraction

(2) Project planning Developing a complex plan through object synthesis requires careful design of the rule base to ensure that steps are carried out in the correct sequence, and that solution failure detection and recovery mechanisms, 7 one of the advantageous features of AI-based design procedures, are correctly interleaved in the overall procedure. Another advantage of the approach is that further knowledge (such as the use of pooled resources 4) may be simply incorporated by adding frames and rule sets as required. (3) Process configuration A 'combinatorial trap' is also possible when considering all the possible competing plans that could be generated during abductive hypothesis assembly. Bylander et al., ~' however, point out that many problems can be solved through enumeration, especially when other means are available to reduce the number of alternative hypotheses formulated. The use of elimination criteria together with acceptance criteria can help reduce the number of partial plans generated. (4) Manufacturing planning and control Developing a consistent set of functionalities is quite a difficult task, even for small models. Again, there are potentially many combinations of the ternary values of input variables: and each corresponding outcome may be meaningful. Tracing and debugging procedures can help validate state abstraction models of this type. Bimson & Burris, 6 for example, outline a consistency checking process for semantic network models which could be useful. A final perspective on the paradigms detailed in this article is provided by Clancey. 14 He avers that 'all knowledge bases contain qualitative models (and) all expert systems are model based'. These qualitative models describe scientific and engineering systems in terms of causal, compositional or subtypicai relationships among objects and events. They thereby occupy an important place alongside quantitative formulations and constitute the essential contribution of AI to science and engineering. In this light, the architectures presented

Knowledge-based systems structure Rules Inference

Environmentsystems Operationsplan Environment & resources Project plan components Plan components Functional plan components

Relationsbetween systems and plan Relations between environment and plan components

Develop plan attributes Develop plan cornponent attributes

Acceptability of plan components Plan component functions

Assembleplan components Activate plan cornponents

here have shown that qualitative modelling and qualitative simulation, combined where required with quantitative procedures, provide a powerful tool for system design and analysis, particularly for complicated systems requiring an integrated and widely-focussed approach.

ACKNOWLEDGEMENT The authors wish to acknowledge helpful material and comments received from B. Chanrasekaran, J. Josephson, E, Eloranta and K.D. Bimson.

REFERENCES 1. 2.

3. 4.

5. 6.

7. 8. 9.

Andreasen, M. M. & Hein, L. Integrated Product Development, IFS (Publications), Bedford, UK, 1987. Augustin, S. & Huebner, H. Designing computersupported production management systems using the aspect system approach, Production Management Systems: Strategies and Tools for Design, H. Huebner (ed.), 1984, Elsevier North-Holland, Amsterdam, Netherlands, 51-66. Avots, I. Application of expert system concepts to schedule control, Project Management Journal, 1985, March. 51-55. Bimson, K. D. & Burris, L. B. Conceptual model-based reasoning for knowledge-based software project management. Proceedings of the Twenty-First Annual Hawaii International Conference of System Sciences, 3, 255 -265. Bimson, K. D. & Burris, L. B, Assisting manager in project definition: foundations for intelligent decision support, IEEE Expert, 1989, Summer, 66-76. Bimson, K. D. & Burris, L. B. Understanding inconsistency in complex semantic networks, LSTC Report, Lockheed Software Technology Center, Austin, Texas, USA, 1989. Brown, D. C. & Chandrasekaran, B. Design Problem Solving, Morgan Kaufman Publishers, San Mateo, Cahfornia, USA, 1989. Burbldge, J. L. The design of production systems, Vlth International Conference on Production Research, Windsor, Ontario, Canada, August 1983. Burbidge, J. L. A classification of production system variables, Production Management Systems: Strategies

Frame-based architectures f o r manufacturing planning and control

based systems in industrial engineering, International Journal of Artificial Intelligence in Engineering, 1990,

and Tools for Design, H. Huebner, (ed.), Elsevier North10. 11. 12.

13.

14. 15.

Holland, Amsterdam, Netherlands, 1989, 3 - 1 6 . Burbidge, J. L. A Production System Variance Connectance Model, Cranfield Institute of Technology, Cranfield, UK, 1984. Bylander, T., Allemang, D., Tanner, M. C. & Josephson, J. R. The computational complexity of abduction, Artificial Intelligence, 1990. Chandrasekaran, B. Generic tasks as building blocks for knowledge-based systems: the diagnosis and routine design examples, The Knowledge Engineering Review, 1988, 3(3), 183-210. Chanrasekaran, B., Tanner, M. C. & Josephson, J. R. Explaining control strategies in problem solving, IEEE Expert, 1989, 4(1), 9 - 2 4 . Clancey, W. J. Viewing knowledge bases as qualitative models. IEEE Expert, 1989, 4(2), 9 - 2 3 . Darwiche, A., Levitt, R. E. & Hayes-Roth, B. OARPLAN: generating project plans by reasoning about objects, actions and resources, Journal of Artificial

26.

27.

28.

29.

30.

16.

17. 18.

19. 20.

21.

22.

23. 24.

25.

5(3), 126-141. Karni, R., and Terry, W.R. Intelligent enhancement of manufacturing planning, Xth International Conference on Production Research, Nottingham, England, August 1989. Karni, R. & Terry, W. R. An industrial engineering viewpoint of engineering design, Industrial Engineering, 1990 (under review). Levitt, R.E. Expert systems in construction: state of the art, Expert Systems for C~vil Engineers, M. L. Maher (ed.), American Society of Civil Engineers, New York, USA, 1987, 85-112. Levitt, R. E. & Kunz, J. C. Using knowledge of construction and project management for automated schedule updating, Project Management Journal, 1985, XVI(5), 57 - 7 6 . Levitt, R. E. & Kunz, J. C. Using artificial intelligence techniques to support project management, Journal of

Artificial Intelligence in Engineering, Manufacturing, 1987, 1(1), 3 - 2 4 .

Intelligence in Engineering, Design and Manufacturing, 1988, 2(3), 169-181. Dattero, R., Kanet, J. J. & White, E. M. Enhancing manufacturing planning and control systems with artificial intelligence techniques, Knowledge-Based Systems in Manufacturing, A. Kusiak (ed.), Taylor and Francis, London, UK, 1989, 137-150. Egilmez, K. & Kim, S. H. A logical approach to knowledge-based control, Journal of Intelligent Manufacturing, 1990, 1(1), 5 9 - 7 6 . Eloranta, E., Syrjanen, M. & Torma, S. Model-based reasoning in manufacturing system design, Knowledge Based Production Management Systems, J. Browne (ed.), Elsevier North-Holland, Amsterdam, Netherlands, 1989, 49-63. Eloranta, E., Syrjanen, M. & Torma, S. Knowledge based tool for manufacturing system design, ComputerIntegrated Manufacturing Systems, 1990, 3(3), 163-170. Even, A. Framework for the development of a hierarchical computer planning and production system, unpublished M.Sc. dissertation, Technion, Haifa, Israel, 1984. Harhen, J. Prospects for AI in production management systems, Knowledge Based Production Management Systems, J. Browne, (ed.), Elsevier North-Holland, Amsterdam, Netherlands, 1989, 3 - 1 3 . Hendrickson, C., Zozaya-Gorostiza, C. A., Rehak, D. R., Baracco-Miller, E. G. & Lim, P. S. An expert system for construction planning, Journal of Computing in Civil Engineering (ASCE), 1987, 1(4), 253-269. Johnson, P. E. & Zualkernan, I. A. Comments on the generic task approach, The Knowledge Engineering Review, 1988, 3(3), 211-212. Josephson, J. R., Chanrasekaran, B., Smith, J. & Tanner, M. C. A mechanism for forming composite explanatory hypotheses, IEEE Transactions on Systems, Man and Cybernetics, SMC-17(3), 1987, 445-454. Karni, R., and Gal-Tzur, A. Paradigms for knowledge-

91

31.

32. 33.

Design and

McClelland, J. L. & Rumelhart, D. E., Explorations in Parallel Distributed Processing, MIT Press, Cambridge, Massachusetts, USA, 1989. Oxman, R. E. Design shells: International Journal of Artificial Intelligence in Engineering, 1990, 5(1), 2 - 8 . Reggia, J. A. & Peng, Y. Modelling diagnostic reasoning: a summary of parsimonious covering theory, Proceedings

of the Tenth Annual Symposium on Computer Applications in Medical Care October 1986, IEEE. 34.

35.

Rolstadas, A. & Hygen, J. Production system variables and influences: impacts on the design of production management systems, Production Management Systems: Strategies and Tools for Design, H. Huebner (ed.), Elsevier North-Holland, Amsterdam, Netherlands, 1984, 139- 143. Sathi, A., Fox, M. S. & Greenberg, M. Representation of activity knowledge for project management, IEEE Tran-

saction on Pattern Analysis and Machine Intelligence, 36. 37. 38.

39.

1985, PAM1-7, 531-552. Sathi, A.. Morton, T. E. & Roth, S. F. Callisto: An intelhgent project management system, AI Magazine, 1986, 7(5), 3 4 - 5 2 . Shaw, M. J. & Menon, U. Knowledge-based manufacturing quality management: a qualitative reasoning approach, Decision Support Systems, 1990, 6(1), 5 9 - 8 1 . Vere, S. A. Planning in time: windows and durations for activities and goals, IEEE Transactions on Pattern Analysis and Machine Intelligence, 1983, P A M I - 5 , 246-259. Vollman, T. E., Berry, W. L. & Whybark, D. C.

Manufacturing Planning and Control Systems, R. D. 40.

Irwin Inc., Homewood, Illinois, USA, 1984. Zozaya-Gorostiza, C. A., Hendrickson, C., and Rehak,

D. R. Knowledge-Based Process Planning for Construction and Manufacturing, Academic Press, San Diego, California, USA, 1989.