You are on page 1of 9

Process planning formalization in an AI framework

Sanjay Joshi, T. C. Chang and C. Richard Liu

School of Industrial Engineerin#, Purdue University, West Lafayette, IN 47907, USA

This paper discusses the process planning problem with a view to automate its solution methods.
The problem is formulated using AI techniques. The basic issues involved in modelling the process
planning problem are discussed. A hierarchical planning structure is proposed, in which the
highest level is a simplification or abstraction of the plan and the lowest is a detailed plan, sufficient
to solve the problem.
States in the problem can be modelled using the CAD representation schemes, such as
Constructive Solid Geometry and Boundary Representation. The advantages and disadvantages
of these schemes are discussed. Operators which correspond physically to material removal are
also modelled using the same representation scheme. The design of a control strategy is discussed.

Key Words: process planning, artificial intelligence, CAD, CAM, solid modelling, hierarchical
planning

1. INTRODUCTION This paper is divided into six sections. Section two


introduces the problem solving techniques in AI. Section
In the area of manufacturing, process planning involves
three gives an overview of Process Planning problem.
the act of preparing a detailed plan for production of a
Existing Computer-Aided Processing Planning (CAPP)
piece part or assembly. The detailed plan contains the
systems are reviewed in Section four. Section five
route, process, process parameters, machines and tools
discusses the proposed AI framework for process
required for production. Process Planning is important in
planning; issues in representations and control strategy
two aspects 1. First it is essential for production; and
are discussed. In the last section a brief summary is given.
second, it determines the efficiency of production.
Artificial Intelligence (AI) methods and principles
provide problem solving techniques which can be used 2. PROBLEM SOLVING IN AI
over a broad spectrum of problem types. Some of these AI
Most of the AI systems show a separation between the
methods exploit knowledge about the problem domain,
standard computational components of data, operations,
and provide some degree of independence between
and control. A production system involves a clean
problem and problem solving techniques2; these AI
separation of these computational components. In
methods are commonly referred to as strong problem
general, the problems in a production system formalism
solvers or knowledge-based systems.
can be represented by the following:
The process of preparing a process plan is mostly
experience based. For example, it is not apparent to 1. an initial state;
everyone what is involved in manufacturing a turbine 2. a goal state;
blade. The interpretation of the engineering drawing, the 3. a set of operators; and
selection of appropriate tools, fixtures and machining 4. a control structure.
parameters, etc., can not be easily modelled mathe-
Transforming the problem statement into components
matically. Many of the decision making rules are based on
of a production system is called the representation
long term experience in manufacturing. Such properties
problem. Usually there are several ways to represent the
make process planning a natural problem for using AI
problem. Virtually any kind of data structure can be used
techniques.
to describe the states. These include symbol strings
This paper discusses some of the available techniques j,n
vectors, sets, arrays, trees, frames and lists. Sometimes the
AI which can be used to model the Process Planning
form of data structure bears close resemblance to some
problem formally. The basic issues involved are discussed
physical property of the problem being solved.
and the proposed framework is explained. An analogy is
One of the frequently used methods used to solve
made between the AI planning systems and the Process
problems in AI is the theorem proving technique 3. Using
Planning system and the similarities discussed.
this technique, we can proceed from the initial state to the
goal state. Proving that the goal state can be reached from
This work was supported,in part by the NationalScienceFoundation the initial state will involve application of operators,
under an EngineeringResearch Center grant #CDR8500022, and a thereby providing a solution to the problem. The
Purdue UniversityDavid RossGrant. application of operators moves the problem from one
AcceptedJanuary 1986.DiscussionclosesSeptember 1986. state to another finally the terminating state. It can either

0267-9264/86/010045-0952_00
© 1986Computational MechanicsPublications Artificial Intelligence, 1986, Vol. 1, No. 1 45
Process planning formalization." S. Joshi, T. C. Chang and C. Richard Liu

move from the initial state to the goal state (forward 3.1 Variant process planning
chaining) or from the goal state to the initial state The variant approach to process planning is based
(backward chaining). upon the concept of group technology. The parts are
Several planning systems have been developed by AI grouped into families and a coding system used to
researchers. For example, in STRIPS 4 a robot problem distinguish each part family. Standard process plans are
solver, the concern is to generate a sequence of moves for created and stored permanently in the database with the
the robot to perform a task, given the task, initial family number as the key. The process planner's job is
conditions, etc. Most planning systems [STRIPS, etc.] then to code each part, classify it into a family, retrieve the
divide the problems that need to be solved into smaller standard plan, and modify it for a particular part. The
pieces and solve these pieces separately. The separate computer assists by providing an efficient system for data
pieces must then be combined to form a single consistent management retrieval, editing and high speed printing
solution to the original problem. This is known as the process plans.
decomposition approach. Decomposition can make Variant process planning systems are best for a
solution of hard problems easier by division of a single moderate number of families each possessing many family
problem into several, hopefully easier, subproblems. If members. In most batch manufacturing systems where
this is not possible then the problem should be divided similar components are produced, variant systems can
into subproblems that have only a small amount of improve planning. Other advantages of variant systems
interaction. are that a variety ofcomponents can be planned using the
For example, consider moving the furniture out of a standard plan once it has been created. Simple
room 2. The problem can be decomposed into a set of programming is required to create a variant system and
smaller easier problems, each involving removing one implementation is easy. However, the variant approach is
piece of furniture out of the room. Within each of these not without its limitations and disadvantages.
subproblems, considerations such as removing drawers Experienced process planners are required to construct,
can be added separately for each furniture. But if there is a maintain, modify, and constantly edit standard process
bookcase behind a couch, then we must move the couch plan for specific components. It cannot be used in an
before we can move the bookcase. These problems are entirely automated manufacturing system where
called nearly decomposable problems. decisions have to be made on line when routings are
Many robot planning systems use the GPS interrupted.
(Generalized Problem Solver) method to generate plans 4.
This process attempts to calculate a difference between
the initial state and final state. The difference obtaining 3.2 Generative process planning
process is performed by a function that is specific for each Generative process planning can be defined as a system
domain of application. The next step in the process which synthesizes a process plan for new components
involves selection of an operator so as to reduce the automatically. It envisions creating a process plan from
difference. GPS works recursively on the preconditions the information available in a manufacturing database
for the operators and when these are satisfied, the without human intervention. It utilizes a computerized
operator is applied to the current state description and the system consisting of decision logic, formulae, algorithms
process continued. Means end analysis is the process of and geometry based data to uniquely determine the
identifying the differences and selecting operators to processing decision for converting a part from rough to
reduce them. finished state. Unlike the variant approach, no standard
Recursively this can be represented as: plans are predefined or stored.
A generative process planning system has two major
generate plan for part ( ); components. The first component is a geometry based
until Si = St input scheme for translating the parts physical features
do and specifications into computer interpretable format.
generate difference between Si and Sr Most existing systems use a coding scheme or part
select an operator to reduce the difference description languages. The second component is the
check if operator preconditions are satisfied software comprised of decision logic to imitate a process
generate plan for subproblem ( ) planner, and compare the part geometry to
update the sate manufacturing capabilities and availabilities of tools and
end. machines. Decisions on process selection, machine
where selection, cut planning, and calculating set up and cycle
Si = initial state; and St= final state. times for each operation are performed automatically. By
Several AI techniques which are applicable to planning making use of decision logic, the process planning
problems are briefly introduced in this section. Their decision making process can be imitated. Decisions on
application to process planning will be discussed irr process selection, sequencing, etc., are made by the system
section five. Before the details are discussed, we will first automatically. The difficulty is transferring component
review current process planning approaches. data and decision rules into computer readable format.
Generative process planning systems have several
advantages. It can generate consistent plans rapidly, new
components can be easily planned and it has the potential
3. OVERVIEW OF PROCESS PLANNING
to be interfaced with an automated manufacturing system.
In developing the factory of the future, automated Most of the current systems are not truly generative.
planning has a vital role to play. The two approaches They need some form of human input, especially to
traditionally used for process planning are the variant describe the part and obtain feature information to drive
approach and the generative approach. the decision based process planning system. The decision

d~ Arti6ri~! l.tolllnpnra !0~6 Vnl I Na ./


Process planning formalization." S. Joshi, T. C. Chang and C. Richard Liu

logic is limited to the ability to check some conditional sequences can be generated from the initial state. Yet a
requirements and select processes. single sequence should be chosen for the final execution.
This final sequence is called a process plan. During the
planning stage, instead of using physical machine tools
4. EXISTING PROCESS P L A N N I N G SYSTEMS
and cutters as operators, a more abstract operator -
Many process planning systems have been developed. A machining process type is used. Machine tools are
typical example of a variant process planning system is the selected later to perform the processes.
CAPP (CAM-I's Automated Process Planning) system In process planning we are concerned mainly with the
developed by CAM-I 5. Other systems are M I P L A N 6, problem of synthesizing a sequence of machining
M I T U R N 7. There are also several prototype systems operations that will achieve the stated goal of producing
developed in academia. the part to required specifications. This is similar to those
There are several generative process planning systems planning problems (e.g., robot problem solving in
used in the industry. For example, A U T O P L A N s, STRIPS) dealt with by AI researchers.
AUTAP 9, GENPLAN*°, XPSII, CPpp12, etc.
Generative process planning systems need a lot more 5.1 Hierarchical process planning structure
research effort to make them truly generative. Several In this paper we propose a hierarchical planning
research systems can also be found, such as APPAS 13, structure for the process planning problem (Fig. 1). A
TIPPS 14, GAR115, P R O P L A N 16, T O M t7, P R P P Is. A hierarchy of plans can be generated in which the highest is
more complete discussion of existing process planning a simplification, or an abstraction, of the plan and the
systems can be found in reference 1. lowest is a detailed plan, sufficient enough to solve the
Both these approaches leave much to be desired in problem. The main advantage of the hierarchical
achieving the goal of total automation. We may add planning structure stems from the fact that it leads to
another classification of process planning - automated distinguishing problem solving techniques critical to the
process planning. An automated process planning system success of the plan and those that are simple details. The
can take the design information directly from the designer method is to first sketch a rough plan that is complete and
then generate a complete and usuable plan automatically. then to refine rough parts of the plan into detailed sub
In order to achieve this goal, AI seems to be a promising plans until finally the plan has been refined to a complete
tool. Although several existing generative systems do use sequence with the required detail to solve the problem.
AI approach, none of them is qualified to be called an Fig. 1 shows a hierarchical structure for process
automated process planning system. To help in attaining planning. At level 1, the design model is decomposed into
this goal, a new framework using AI techniques is volumes for machining. The volume decomposition can
proposed to formulate the problem. be obtained by feature recognition using seman-
tics 19"20"21'22 o r by algorithmic decomposition 2a'24.
5. PROCESS P L A N N I N G AND AI F R A M E W O R K Level 2 involves planning the machining of each volume
by determining the sequence of machining operations. At
In order to provide a formal description of the process
planning problem it is necessary to
1. Define a state space that contains the possible Part Design
configurations of the relevant objects. t
Level 0
2. Specify one or more states within that space that
describe possible situations from which the problem develop plan for part
solving process may start. These states are the initial

3.
states.
Define goal states, which are one or more states
I
Rough Plan (general
acceptable as solutions to the problem. operation categories
4. Specify the set of operators or actions available.
5. Develop an appropriate control strategy, which will
determine the order in which to apply the operators. Level 1 decompose into volumes 1
for machining
In the part manufacturing problem in which process
planning is to solve, the initial state is the raw material or l
Decomposed Volumes
workpiece from which the part is to be produced, and the
goal state is the final part. The set of operators comprises
½
the available machine tools and cutters. The process of Level 2 plan the machining of each
producing the part involves proceeding from the raw volume by determining
which machine operation or
workpiece (initial state) to the final part (goal state). set of machining operation
Application of the operators (the machine tools) moves will be used
the problem from one state to another. The various stages
represent the workpiec¢ in 'work in progress' condition.
l
Process Sequence
S,i,~, --, Sl --' $2 - , . . . ---, Sr,,i 1
The states are non-reversible, i.e., the application of an Level 3 for each machining operation
determine specific machine
operator cannot return to a previous state, this is because and cutting parameters, feed,
the material once removed cannot be added back. speed, etc.
However, in the case of process planning, the actual
material removal has not been done, alternative process Fig. I. Hierarchical structure for process planning

Artificial lntelliaence, 1986, Vol. 1, No. 1 47


Process planning formali:atiott" S. Joshi, T. C. Chang and C. Richard Liu

knowledgebase
(theoremand rules)
interactive
interface
program _1 p obl m I
1 J interpretation l
and
w i t h geometric
"-[ descriptionJ -I I -I implementation I
7--

mode.tiers
T
Fig. 2. Division of the process planning problem (taken from Ref. 30)

I
DRILL ~

Fig. 4a. Effect of operator DRILL


Fig. 3. Slotted hole to be manufactured

this stage more details are taken into account, such as the
desired surface characteristics. Some expert systems have
been developed to assist in planning machining, for
example GARI15, TOM17, EXCAp25, Nau and
Chang 26"27. Level 3 deals with determining the cutting
feed, speeds, depth of cut etc. Systems using expert
j BORE ~ t

I
knowledge have been developed to assist at this level. Fig. 4b. Effect of operator BORE
Examples are CUTTECH 2s, XCUT 29. In this paper,
several issues in problem representation and control at
these levels are discussed. However, our current work is The effect of the operator can be represented as shown in
focussed mainly on level 1, since it is at this level that the Fig. 4a.
CAD and CAM are interface and currently this is the part
which needs to be automated. (ii) Operator: BORE
precondition:
5.2 Problem decomposition hole exists
The decomposition approach discussed in Section 2 the surface to be made is hole or a slot
can also be applied in solving process planning problems. The effect of this operator can be represented as shown in
This can be illustrated by Fig. 2 30. The process planning Fig. 4b.
problem is nearly decomposable, similar to the furniture An approach similar to GPS can be used to assist in the
removal problem. For example, to produce the slotted
development of process plans. Applying the GPS
hole shown in Fig. 3, it can be decomposed into two
procedure (Fig. 53, the difference between So and Sr is A 1.
subproblems. One to machine the hole and the other to To reduce the difference, it can be seen that the operator
machine the slot. The problem is not completely BORE can be selected, but since the precondition and So
decomposable, since there is a constraint between the two has difference A 2, between them, so bore cannot be
divisions. The hole has to be made before the slot. An applied directly. Checking to see if the difference A2 can be
approach similar to GPS can be used to assist in the reduced, operator DRILL can be applied, since there is no
development of process plans for manufacturing. difference between So and the precondition for DRILL.
Consider the example shown in Fig. 3. The initial state Applying operator D R I L L implies updating the state
is a raw material blank, and the final state is the blank with from So to $I. Now again going through the procedure,
the slotted hole. The operators are we can apply the operator BORE to obtain the final part.
(i) Operator: DRILL Thus the reverse sequence in which the operators are
precondition: applied gives the process plan.
raw material exists This example is an abstraction of the real problem, and
the surface to be made is a hole provides a way of separating important problem features

48 Artificial Intelliaence. 1986, Vol. 1, No.-1


Process planning formalization." S. Joshi, T. C. Chang and C. Richard Liu

I The CSG method uses Boolean combination (union,


S0 ~ ~ S! intersection, or difference) of simple solids called
primitives to build a part. The construction procedure is
stored internally as a tree structure, with nodes
I
|
representing regularized set operators and the leaves are
primitive solids. CSG schemes are not unique. There are
many possible representations for a single object. CSG
trees are concise and easy to create, and their validity can
Difference ZXI~ be easily checked.
BREP is a method of representing a solid object using
its boundary surfaces. The faces of the solid are
represented by its boundary edges, and vertices. BREP
schemes are unambiguous, and generally unique. To
I establish validity of BREP schemes both geometry and
topology have to be verified. In a BREP model the
information between the faces, edges, and vertices and
Difference A2 between preconditionof B O R E a, So their relationships is easily available.
The choice of a representation scheme for the purpose
of developing an automated process planning system to
link CAD and CAM makes part representation a major
consideration.

5.4 CSG vs BREP in mant(acturing applications


I CSG schemes based on simply solids as primitives seem
to be most suited for manufacturing applications.
Features of the parts such as holes, slots, and so forth may
T h e sequence to obtaln the final state Is actually be present explicitly in the tree. An example of
explicit representation of a feature, is that the subtracting
of a cylinder can create a hole. The definition of the hole
DRILL
can be obtained from the definition of the cylinder. Thus
BCRE
the information can be used directly for generating
manufacturing process plans, since the subtraction of the
cylinder can be associated with the drilling operation.
Unfortunately, this is not always the case.
Fig. 5. Application of the GPS procedure A CSG scheme is not unique, and a part constructed by
subtraction operations may also be created using the
union operations. The union operator has no
from others that will overwhelm the process. It serves to corresponding material removal operation. In such cases
illustrate the idea, and opens up various issues needed to the features to be machined cannot be obtained directly
achieve the desired goal. In the forth coming sections the from the tree, but by evaluating it.
issues are highlighted and discussed. CSG trees may also contain irrelevant and redundant
information. Design iterations during the design phase
5.3 Representation of states for process planning introduces elements into the CSG tree which may not
One of the foremost issues is 'How should we represent even exist in the final part. Algorithms for application
the various states?'. A good system for representation of programs using such models must discover and discard
complete structural knowledge in a particular domain irrelevant information by evaluating the tree. This can be
should possess four properties 2. It should adequately a costly and time consuming operation especially if the
represent all kinds of knowledge needed in the domain. tree is large.
The representation should allow manipulation of In a BREP, the model is represented explicitly and the
structures in such a way as to derive new structures features are represented implicitly. The faces, edges and
corresponding to new knowledge inferred from old. It vertices are represented explicitly as they are in the final
should be efficient in its ability to incorporate into the part; since boundary surfaces are what one sees when
knowledge structure additional information that can be looking at an object it seems as a reasonable approach to
used to focus the attention of the inference mechanism. object definition. For a planar object there is no problem
Finally, it should have the ability to acquire new when decomposing into constituting faces. However, for an
information easily. There are several ways to represent a object shown in Fig. 6, it is difficult to visualize the faces
problem. bounding the object. Several different ways can be used to
For the purpose of automating process planning, the decompose the object face into smaller faces for clarity.
best representation of the states would be the one used in
BREP seems directly usuable for CAM. Many
geometric modelling systems. This would enable
manufacturing processes deal with surfaces, e.g., machining
automatic generation of plans from the CAD data. The
operations cut surfaces. Edges and vertices emerge as a
most commonly used representations used in geometric
consequence of meeting of surfaces and cannot be
solid modelling systems are 31.
manufactured independently. The information required
1. Constructive Solid Geometry (CSG) for manufacturing, such as the calculations of tool paths,
2. Boundary Representation (BREP) can be extracted easily from an explicit surface definition.

Artificial lntelliaence, 1986, Vol. 1, No. 1 49


Process planning formalization: S. Joshi, T. C. Chang and C. Richard Liu
are compound features. Some systems have been built
using this approach, for example, X C U T 29 and the NBS
system35.
The method would be suitable if the DSG tree can be
constructed from a design CSG tree. Using the AI
approach of selecting operators and applying these
operators, a DSG can be constructed. It corresponds to
creating a volume decomposition of the part to be
manufactured.
5.6 Inadequacies of solid modellers
Tolerancing and surface finish which is an integral part
of the specification of mechanical parts is lacking in
current solid modellers. This is a serious deficiency 36. For
fully automated process planning tolerancing infor-
Fig. 6. Object with difficulty in visualizing bounding faces mation is needed. Current generation solid modellers
(taken from Ref. 31) contain complete representation of nominal parts. The
tolerancing information is added on latter for process
planning.
The necessity for identifying surfaces in a final model of
5.7 Representation of operators
design objects for CAM applications forces systems based
on other solid modelling schemes to perform a conversion The operators in the process planning problem are the
machining or metal removal operations. A sufficient
of BREP at some time before they can be used for CAM
collection of operators is needed so as to be able to
purposes.
convert the initial state through many intermediate states
In the modelling of manufacturing process the material
to the goal state.
removal can be looked upon as removal of volumes to
Since the operators will result in change of state, the
create new surfaces a2. Hence, it would need the ability to
representation of the operators should be such so as to
convert from CSG to BREP and vice versa. Converting
allow easy modification of the states. The states are
fromCSG to BREP is not difficult, but converting from
BREP to CSG is non trivial 31. Nau et al. 33 discuss a represented in the geometric solid modellers, so the
conversion method for converting from BREP to CSG. operators should also be represented in the solid modeller.
The proposed system will select BREP as the The operators physically correspond to material
representation. Algorithms will be developed to removal operations and imparts certain surface
characteristics to the new surfaces generated. The
decompose the part to be machined. Combined with the
complete modelling of the operators must take into
knowledge on manufacturing practice, machinable
account several factors such as:
surfaces can be identified.
1. type of the machine tool used;
5.5 Other representations for process planning 2. any preconditions needed for the application of the
applications machine tool;
With a view to achieving a CAD representation of an 3. type of relative motion between the tool and the
object suitable for CAM operations Arbab 34 proposed workpiece;
using a Deforming Solid Geometry (DSG). DSG is a 4. tool geometry;
'methodology for describing mechanical parts and 5. generated surface characteristics; and
assemblies through a sequence of operations resembling 6. range of motion.
those of manufacturing'. DSG's approach to the The above mentioned factors can be used to determine
modelling of material removal operations is in principle the configuration of the next state. The details required for
not far from that in CSG. The starting point in DSG is the modelling of the operators depends on the level of planning
initial workpiece. Material removal operations are in the hierarchy. For example, at level 2, the operator can
characterized by the fact that they can be associated with be written using shape codes or descriptions.
predetermined geometric shapes. A 'tool' in DSG is a
specification of material removal operation. Using the IF
operator 'apply' and the 'tool' yields another workpiece the volume to be removed has a X-shape
by removal of material. Subsequent applications yield the and surface finish greater than Y,
final workpiece. THEN
The major problem with such an approach stems from chose Process-Z.
the fact that the designer will have to build in th~ However, at level 1, both I F and T H E N portion of the rule
manufacturing logic, during the design phase, forcing the are represented in geometric modeller. A simplified
designer to be an expert at manufacturing also. It imposes operator can be written as:
an operational structure and discipline which can restrict IF
creativity of the designer and also imposes unnecessary F o is an entrance surface
restriction on the flexibility of the CAD system. It is not and four adjacent surfaces F 1, F 2, F 3 and F 4 are wall
clear how the DSG will solve the problems arising due to surfaces
design iterations. Usually another approach associated and a bottom surface F 5 exists
with the approach is the use of macro process plans and THEN
part programs for each 'tool' (feature). Macros do not remove the volume enclosed by the surfaces
always generate a feasible final plan especially when there and call it a pocket.

50 Artificial Intelligence, 1986, Vol. 1, No. 1


Process planning formalization: S. Joshi, T. C. Chang and C. Richard Liu
Entrance surface, adjacent surface, wall surface and of machining process can be modelled in geometric
bottom surface will have to be defined. modellers 38.
Sugimura et al. 37 in a paper discuss the formulation of At this level we only look at the geometry of different
shape generation process of machine tools. The model is operations. For example a turning operation will produce
described by the relative motion between the workpiece a cylindrical surface, or a tapered cylindrical surface, a
and the cutting edge with a specified geometry. The drill will produce an internal cylindrical surface, a milling
different types of cutting motion are linear and rotary and shaping operation will generate a plane surface, etc.
cutting motion. The relative motion of the cutting edge This information can be used along with the existing
and workpiece is obtained by the feed motion of the condition of the workpiece state to assist the
workpiece and/or tool motion. decomposition and solving the subgoals obtained, in an
At the level of abstraction being considered at this stage effort to reach the final goal.
(level 1), the volume of material removed can be obtained
by considering the direction of feed and type of cutting
motion. This can be obtained in a geometric modeller by 5.8 Control strategy
the sweeping action. Sweep is of two types, translational The extent to which an operator is applied and the
and rotational 31. It is a technique for forming a solid logical structure to decide in which order to try the
object by moving a 2-D shape along a trajectory. operators is called control strategy.
Translational sweeping involves a trajectory along a line At any instant, the order of trying and applying the
segment, where as in rotational sweeping the trajectory is operators itself is dependent on the then current state of
a circular arc (Fig. 7). The concept of sweep can be the problem. During the control structure design, an
extended to sweep of 3-D objects. Shape modifying effects evaluation function, which may represent a weighted cost
solution path can be used to evaluate the solutions
Z obtained.
The problem with process planning is that the subgoals
are not independent, but are conjunctive, i.e., they
SWEEP~ A interact. They cannot be achieved in an arbitrary order.
The problem of constructing an efficient plan for
achieving conjunctive goals that are not independent can
be approached by the methods described below.
1. Formulate plans that have interactions between
subgoals and then fix them by re-ordering the
problem solving operators in the plan. This method is
used by H A C K E R 39 and INTERPLAN 4°.
2. Another method used by Waldinger 4t is more
constructive. It does not re-order operators in a
flawed path, but develops a plan by inserting
operations one by one, checking each for potential
interference with established operations.
3. A useful constraint handling technique to effectively
manage interactions between subproblems in a
hierarchial planner is described in 4z.
The use of means end analysis reduces the amount of
search done by the problem solver. It recognizes only one
type of goal, namely, to reduce the difference between the
,X two states. The problem solving operators can be
Z classified according to the kinds of difference they reduce.
Consequently, only a fraction of the available operators
will be applicable to any given goal, and search among the
operators for an applicable one will be reduced.

A--_J ~SWEEP There may be several applicable operators and an a


priori basis for selecting one. Also, it may not be possible
to know whether the subgoals of an operator will be
satisfied or if the evaluation will lead to a dead end. The
control strategy then has to back track to find an
alternative path. In a complicated problem, there may be
several alternate paths, and a substantial amount of back
X T

Y
tracking may be needed to solve the problem.
In the process planning problem, several operators may
be applicable, but the choice may be restricted by several
factors which are considered good manufacturing
practices. For example, consider reducing the difference
shown in Fig. 8.
B Two operators, drilling and milling are applicable.
Fig. 7. Translational and rotational sweep (taken from Using the drilling operator before the milling operator
Ref. 31) results in the difference shown in Fig. 9 and the milling

Artificial Intelliaence. 1986. Vol. 1. No. 1 51


/ /
Process planning formalization." S. Joshi, T. C. Chang and C. Richard Liu

Difference to be reduced

I I
j Ist Application

2rid Application

Fig. 8. Difference to be reduced


2rid Application
ot MILLING NN

Ist, Application
of MILLIN(3
,, Fig. 9. Effect o f using D R I L L b e f o r e M I L L I N G Fig. I0. Different ways of applying the M I L L I N G
operator

operaror has to applied to a volume with a hole in it, directly for process planning, since the domain is too
which is not considered good manufacturing practice. limited. An integration of the two is necessary.
Applying the drilling operation after the milling is better This paper presented a formalization of the process
practice. This induces an ordering of the operator for this planning problem in an AI framework. Several issues
particular case. involved in representation, control strategy and
Further, there may be several ways to reduce the modelling of the operators have been discussed, with
difference using the same operator in different ways; for reference to various levels of the planning hierarchy
example, to reduce the difference described above, the described in Fig. 1. The problem of generating process
milling operator can be applied twice as shown in Fig. 10. plans automatically needs a lot more research and
Different methods may be suitable for different detailed implementation needs to be looked at. An
situations or cases. The choice of the method to use should important issue that was not addressed is the integration
be determined by the control strategy. Since the of several existing knowledge-based systems and
application of the operators in a process planning system algorithmic programs to provide an unified framework
is governed by good manufacturing practices, these can be for a fully automated process planning system. The
incorporated in a rule based system to assist in selecting Blackboard architecture described in (Ref. 43) seems to
an operator. have potentials for such an integration.
It is also worthwhile to note that the evaluation
function is used to assist in the selecting of operators. By
appropriately selecting the cost factors desirable ACKNOWLEDGEMENTS
processes will be considered over other processes. For
Authors would like to express their appreciation to the
example, a twist drilling process will be selected instead of reviewers and the editm' for their valuable suggestions.
an end milling process for producing a round hole because
in general, the former costs much less. There is no
intention of finding an 'optimal' process plan, since a plan REFERENCES
exists only after the final schedule is set. A goal is to find a
'good' process plan. f Chang,T. C. and Wysk, R. A. An Introduction to Automated
Process Planning Systems, Pren/ice Hall, Englewood, New
Jersey. 1985
6. D I S C U S S I O N 2 Rich, E. Artificial Intelligence, McGraw-Hill, New York, 1983
3 Green, C. Application of Therom Proving to Problem Solving,
O u r previous experiences with building process planning IJCAI l, 1969, 219-239
systems indicates that for the success of a completely 4 Nilsson,N. Problem Solving Methods in Artificial Intelligence,
a u t o m a t e d system a unified approach is necessary. McGraw Hill, New York, 1971
5 Link, C. H. CAPP - CAM-I Automated Process Planning
Several systems have been built - GARI, T O M , System, Proceedin#s of 13th Numerical Control Society Annual
P R O P L A N , T I P P S , SIPP, C U T T E C H , X C U T etc. - but Meeting and Technical Conference, Cincinnati, Ohio, 1976
these are limited to process selection and planning the 6 Schaffer,G. GT via Automated Process Planning, American
machining operations. Part interpretation, a major issue, Machinist, May 1980, 119-122
7 TNO, Introduction to MIPLAN, Organization for Industrial
is performed manually by converting the part description Rsearch, Inc., Waltham, Mass., 1981
into some specialized input format for these systems. Part 8 Tempelhof,K. H. A Systemof Computer Aided ProcessPlanning
interpretation techniques developed are not suitable fo~'Machine Parts, SME Technical Paper, Series MS79-154, 1979

52 Artificial Intelligence, 1986, Vol. 1, No. 1


Process planning formalizatiotu S. Joshi, T. C. Chang and C. Richard Liu

9 Evershiem, W. and Esch, H. Automated Generation of Process Process Planning, Annals of CIRP, 1984, 33/1/84, 303-306
Plans for Prismatic Parts, Annals of CIRP, 1983, 32/1/83, 361- 26 Nau, D. S. and Chang, T. C. Prospects for Process Selection
364 Using AI, Computers in Industry, 1983, 4, 3. 253-263
10 Tulkoff, J. Lockheed's GENPLAN, Proceedings of 18th 27 Nau, D. S. and Chang, T. C. Hierarchial Representation of
Numerical Control Society Annual Meeting and Technical Problem Solving Knowledge in a Frame Based Process Planning
Conference, Dallas, Texas, 198 I, 417--421 System, Journal of Intelligent Systems, to appear
11 CAM-I, Functional Specification for an Experimental Planning 28 Barkocy, B. E. and Zdeblick, W. J. A Knowledge Based System
System XPS-I, Computer Aided Manufacturing - International, for Machining Operation Planning. Proceedings of
Arlington, Texas, 1979 AUTOFACT 6, pp. 2.11-2.25, 1984
12 Dunn, M. S. and Mann, S. Computerized Production Process 29 King, M. S., Brooks, S. L. and Schaefer, R. M. Knowledge Base
Planning, Proceedings of 15th Numerical Control Society Annual Systems: How will They Affect Manufacturing in the 80"s?,
Meeting and Technical Conference, Chicago, 1978 Technical Report BDX-613-3185, Bendix Kansas City Division,
13 Wysk, R. A. An Automated Process Planning and Selection 1985
Program: APPAS, PhD Thesis, Purdue University, West 30 Liu, C. R. and Liang, G. R. Logic Approach to Automatic NC
Lafayette, Indiana, 1977 Program Generation, Lecture Notes, Intelligent Manufacturing
14 Chang, T. C. and Wysk, R. A. Integrating CAD and CAM Systems, International Summer Seminar, Yugoslav Committee
Through Automated Process Planning, International Journal of for Electronics and Automation, Dubrovinik, Yugoslavia,
Production Research, 1984, 2, 5, 877-894 September 1985
15 Descotte, Y. and Latombe, J. C. GARI: An Expert System for 31 Requicha, A. A. G. Representations for Rigid Solids: Theory,
Process Planning. In Solid Modeling by Computers, edited by M. Methods, and Systems, Computing Surveys. 1980, 12, 4, 437--464
S. Pickett and J. W. Boyse, Plenum Press, New York, 1985 32 Joshi, S. Volume Deeompostion Algorithms for Process
16 Mouleeswaran, C. B. PROPLAN: A Knowledge Based Expert Planning, PhD Thesis Proposal, 1985
System for Manufacturing Process Planning, Master's Thesis, 33 Nau, D. S., Jones, D. and Masai, K. Converting Boundary
University of Illinois at Chicago, 1984 Surface Representations into Constructive Solid Geometry
17 Matsushima, K., Okada, N. and Sata, T. The integration of CAD Representation, IEEE Transactions Systems, Man and
and CAM by Application of Artificial Intelligence Techniques, Cybernetics, 1985, to appear
Annals of CIRP, 1982, 31/1/82 34 Arbab, F. Requirements and Architecture of a CAM Oriented
18 Liu, C. R. and Srinivasan, R. Generative Process Planning - A CAD System for Design and Manufacture of Mechanical Parts,
Syntactic Pattern Recognition Approach, Computers in PhD Thesis, University of California, Los Angeles, 1982
Mechanical Engineering, 1984, 2, 5, 63--66 35 Brown, P. F., Mclean, C. R. and Whitt, N. M. Data Preparation:
19 Choi, B. K., Barash, M. M. and Anderson, D. C. Automatic Process Planning, Technical Briefing Notes, National Bureau of
Recognition of Machined Surface from a 3-D Solid Model, Standards, 1985
Computer aided Design, 1984, 16, 2, 81-86 36 Requicha, A. A. G. Representation of Tolerances in Solid
20 Henderson, M. R. Extraction of Feature Information from Three Modelling: Issues and Alternative Approaches. In Solid Modeling
Dimensional CAD Data, PhD Thesis, Purdue University, West by Computers, edited by M. S. Pickett and J. W. Boyse, Plenum
Lafayette, Indiana, 1984 Press, New York, 1985
21 Wolfe, P. M. and Kung, H. K. Automating Process Planning 37 Sugimura, N., lwata, K. and Fuminori, O. Formulation of Shape
Using Artificial Intelligence, Proceedings of Annual lnternation Generation Process of Machine Tools, Proceedings of the
Industrial Engineering Conference, Institute of Industrial IFAC'81, Kyoto, Japan, 1981, XIV, 158-163
Engineers, 1984, 387-395 38 Hunt, W. A. and Voelcker, H. B. An Exploratory Study of
22 Kyprianou, L. K. Shape Classification in Computer Aided Automatic Verification of Programs for Numerically Controlled
Design, PhD Thesis, Christ College, University of Cambridge, Machine Tools, Technical Report TM-34, University of
Cambridge, UK, 1983 Rochester, Rochester, NY, 1980
23 Grayer, A. R. The Automatic Production of Machined 39 Sussman, G. J. A Computational Model of Skill Acquisition, AI
Components Starting from a Stored Geometric Description. In Technical Report, Rep. 297, AI Lab., MIT, 1973
Advances in Computer Aided Manufacturing, edited by D.' 40 Tate, A. InteractingGoalsandtheirUse, IJCAl4, 1975,215-218
McPherson, North-Holland Publishing Co., Amsterdam, 1977, 41 Waldinger, R. Acheving Several Goals Simultaneously. In
137-151 Machine Intelligence 8, edited by E. W. Elcock and D. Mitche,
24 Woo, T. C. Interfacing Solid Modelling to CAD and CAM: Data Wiley and Sons, New York, 1977
Structures and Algorithms for Decomposing a Solid, Technical 42 Stefik, M. Planning with Constraints (MOLGEN 1), Artificial
Report 83-6, Department of Industrial and Operations Intelligence, 1981, 6, 11-140
Engineering, University of Michigan, Ann Arbor, 1983 43 Hayes-Roth, B. A Blackboard Model of Control, HPP 83-38,
25 Davies, B. J. and Darbyshire, I. L. The Use of Expert Systems in Dept. Computer Science, Stanford University, August 1984

Artincial lntelliaance_ IgR6_ l,"nl. I No. I 53

You might also like