International Journal of Production Research, 2005, in Print

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 14

International Journal of Production Research, 2005, in print

Variant process planning of castings


using AHP-based nearest neighbour algorithm for case retrieval

R. G. Chougule, Research Scholar,


Mechanical Engineering Department,
Indian Institute of Technology, Bombay
chougule@me.iitb.ac.in

B.Ravi, Associate Professor,


Mechanical Engineering Department,
Indian Institute of Technology, Bombay
bravi@me.iitb.ac.in

ABSTRACT

This paper presents our research work on developing a computer-aided casting process-planning system
using case based reasoning. The nearest neighbour algorithm has been used for retrieval of a previous case
(solution) that is closest to the current case (problem) under consideration. The algorithm is driven by
product attributes related to geometry (size, shape complexity, section thickness, etc.), quality (surface
finish, tolerance, maximum void size) and production (order quantity, production rate, lead time). Shape
complexity, which plays an important role in process planning, has been quantified based on geometric
parameters of the casting model. Weights to attributes have been determined using analytic hierarchy
process (AHP), which allows pair-wise comparison and consistency checking of judgments. The complete
methodology has been implemented in a web-based framework to enable early manufacturability
assessment of castings through cost estimation and process simulation, both of which require process
planning information. An industrial example is presented to illustrate the methodology.

Keywords: Analytic Hierarchy Process, Case Based Reasoning, Casting, Computer Aided Process
Planning, Design for Manufacturability, Shape complexity.

1. INTRODUCTION

Process planning involves determining the process steps, process parameters and instructions to
manufacture a part economically and with acceptable quality, taking design information as input (Zhang
and Alting 1994). Computer aided process planning (CAPP) has become an essential element of computer
integrated manufacturing systems, since it links design and manufacturing activities, and enables concurrent
engineering. Investigations have shown that CAPP systems could result in reduction of manufacturing cost
by up to 30%, and the manufacturing cycle time and total engineering time by up to 50% (Younis and
Waheb 1997). The term process planning has however, come to imply machining process planning since
most of the research has been conducted in this area, as reflected in review papers covering technical
literature on CAPP (ElMarghy 1993, Zhang and Alting 1994).

Metal casting is group of manufacturing processes, in which molten metal is led into a dispensable or
permanent mould, and the solidified casting is subsequently machined to obtain the final shape. Cast parts
are found in 90% of manufactured goods and equipment (DoE, 1999). Major families of casting processes
include sand casting, investment casting, die casting and centrifugal casting. Each of these in turn has a
number of individual processes, depending on the type of tooling, moulding, core-making, melting and
pouring practice. For example, sand processes include green sand, shell mould, sodium silicate, hot box,
cold box, etc. Green sand casting can be further classified as manual, machine moulding, high pressure
moulding and Disamatic (vertical) moulding, to name a few. Each of these processes is characterized by
different tooling materials, process parameters and resources consumed. Selecting a suitable casting process
and determining its process parameters is thus a challenging task, requiring a detailed knowledge of various
processes and experience in process planning for a variety of products. However, existing CAPP systems
(which are primarily meant for machining process planning) consider a cast part only as a raw material, and
do not have any facilities for planning the casting process prior to machining. Since process planning
directly affects the quality and cost of a cast product, there is a need for developing a CAPP system for cast
parts. This has been taken up in the present investigation. An additional challenge is in driving the process

1
planning by product data, enabling design for manufacturability through process simulation (Chougule and
Ravi 2004a) and early cost estimation (Chougule and Ravi 2004b).

The paper starts with a review of literature related to casting process planning, process planning approaches
in general and variant process planning in particular, and case based reasoning methodology. This is
followed by our proposed methodology for retrieving a suitable previous case using a nearest neighbour
algorithm based on analytic hierarchy process (AHP), which is rapidly emerging as a powerful tool for
multiple-criteria decision-making. Finally, the implementation of the proposed methodology for computer-
aided casting process planning is described and illustrated with an industrial example.

2. PREVIOUS WORK

Earlier, Ajmal and Dale (1987) developed a simple computer aided process planning and cost estimating
system for foundry application driven by a data base and interactive user input. However, most of the
reported work in the domain of casting process planning focuses on casting process selection, which is only
the first step towards process planning. Sirilertworakul et al. (1993) developed a knowledge base for alloy
and process selection for casting. Darwish and El-Tamimi (1996) developed a preliminary casting process
selection expert system (PCPSES) using Rule Master, a software package for developing expert systems. Er
et al. (1996) developed a knowledge-based expert system for process selection for cast components. Akarte
et al. (1999) used Analytical Hierarchy Process methodology for casting process selection.

In the machining domain, process planning has been a major area of research for the last two decades.
There are two main approaches: generative and variant. A number of process planning systems using these
approaches have been reported in literature (Zhang and Alting 1994, Gu and Norrie 1995). The generative
approach involves automatic generation of process plan based on manufacturing information stored in the
database, decision-making logic and algorithms. This requires a comprehensive knowledge base, which is
difficult to develop and maintain. On the other hand, variant process planning is based on retrieval and
adaptation of a standard process plan stored in a database. This is easier to implement in practice, and has
received more attention from researchers so far. Semi-generative approaches combine both approaches
(Markus et al. 1997); the variant approach is used to identify a previous case, and a generative approach is
used to adapt the previous case to the present problem using a knowledge base in the form of if-then rules.
Two approaches for variant process planning: group technology and case based reasoning are briefly
reviewed here.

2.1. Variant process planning

The variant process planning involves grouping the parts into families and storing a standard process plan
for each family, which is then retrieved depending on the family of a new part and then modified manually.
Several Group Technology (GT) coding schemes such as OPTIZ, MICLASS, COFORM, KK-3 (Optiz
1970, Chang and Wysk 1985) have been used for part classification and coding, especially for machining
process planning. Yaskovskii et al. (1971) proposed a seven digit code number for classification of castings
based on alloy grade, batch size, weight, complexity, production process and equipment. A similar system
has been proposed by Law et al. (1981) based on shape, complexity, weight, material, quality, and size.
Creese (1979) stressed the need of group technology for casting. However, the large variety of castings (in
terms of parameters related to shape, metal, process and quality) and an even larger variation in their
process plans appears difficult to capture by a limited digit code structure. Another limitation of group
technology is the effort required for classification and coding. These limitations can be overcome by case
based reasoning.

2.2. Case based reasoning

The case based reasoning (CBR) is an analogical reasoning method, which relates some previously solved
problem or experience to a current unsolved problem to form analogical inferences for problem solving
(Aamodt and Plaza 1994). It involves: (a) storing previous cases (projects) in a case base, (b) retrieving a
similar previous case based on a set of product attributes, and (c) adapting the solution (process plan) of a
previous case to the new case. The CBR methodology has become popular in the last few years because of a
number of benefits involved such as finding the solutions to complex problems quickly, capturing the
experience of a skilled specialist and discovering decision knowledge in hidden data (Bergmann et al.
1998). The case based process planning systems have been reported for prismatic components (Marefat and

2
Britanik 1997, Tiwari et al. 2001), machining of cylindrical components (Chang et al. 2000), and forged
components (Lei et al. 2001).

The success of a CBR system depends on its ability to retrieve the most relevant case in support of solution
to a new case. A number of methods are available for case retrieval such as induction, knowledge based
indexing and nearest neighbour algorithm (Kim and Han 2001). The nearest neighbour algorithm has
emerged as the most popular method. It involves specifying the values and weights of attributes, based on
which the nearest case (neighbour) is identified and retrieved. For identifying the nearest neighbour,
similarity of problem case (target) to a case in library (case base) is determined by:
n
Similarity (T , S ) = ∑ f (Ti , S i ) × wi (1)
i =1
where,
f - Function to determine distance between attributes of target case T and source case S
n - Number of attributes in each case
i - Individual attribute from 1 to n
w - Weight of attribute i

The domain knowledge of an expert is important for setting the weights (Kolodner 1991). Wettschereck
reviewed several weight setting methods such as incremental hill climbers, continuous optimisers,
conditional probabilities and class projection to derive weights (Wettschereck et al. 1995). Weight setting is
difficult if the number of attributes are large. This problem can be overcome by Analytic Hierarchy Process
(AHP), in which only two attributes are compared each time to calculate the weights (Satty 1994). The main
steps in AHP involve structuring the problem as a hierarchy of overall objective (goal) followed by sub-
goals, criteria and sub-criteria; pair-wise comparison to determine the dominance of one element over the
other element (please refer the appendix); weight (priority) calculation and checking the consistency of
comparison. The AHP provides a useful and systematic methodology for determining the attribute weights,
as well as check the consistency of judgments during pair-wise comparisons.

To summarise, there is a large amount of technical literature on machining process planning, but very little
work has been reported in casting process planning. Over the last few years, case based reasoning
methodology has become popular for process planning since it captures a richer picture of the process and
provides automated mechanisms for retrieving a similar past case that can be adapted to a new case. This
motivated us to develop a systematic casting process planning system using case based reasoning
methodology. The methodology has been implemented in a 3D CAD environment to facilitate automatic
computation of geometry related attributes and enable design for manufacturability.

Figure 1. Overall casting process planning methodology

3
3. CBR METHODOLOGY FOR CASTING PROCESS PLANNING

The methodology comprises the following steps, which are explained in subsequent subsections (figure 1).
• Identifying the attributes of a casting for case retrieval
• Determining attribute weights using analytic hierarchy process
• Specifying the attribute values
• Retrieving one or more similar cases from the case base
• Adapting the selected case to handle the new problem and retaining the adapted case for future use.

3.1. Casting attributes

Determining the product attributes that describe a new problem completely and correctly is an important
step in the CBR methodology. Fifteen attributes have been identified for casting process planning. These
include: casting material, maximum casting size, casting weight, minimum and maximum section thickness,
minimum and maximum core size, shape complexity, dimensional tolerance, surface finish, maximum void
size, order quantity, production rate, sample lead time and production lead time. These play an important
role in process and equipment selection for manufacturing the casting and have been identified based on
discussions with expert casting engineers, product designers and purchase managers.

3.2. Attribute weights

For computing the attribute weights, analytic hierarchy process has been employed. As mentioned earlier,
the AHP methodology involves developing hierarchical structure of the decision problem, determining the
relative priorities of attributes by pair-wise comparison, checking the consistency of pair-wise comparisons,
and calculating the weights based on relative priorities.

3.2.1. Structuring the attributes

All the attributes (except casting material) are structured in a hierarchal form as geometry related attributes,
quality related attributes and production related attributes as given below (figure 2):
• Geometry related attributes: maximum casting size, casting weight, minimum and maximum section
thickness, minimum and maximum core size, and shape complexity
• Quality related attributes: surface finish, dimensional tolerance and maximum void size
• Production related attributes: order quantity, production rate, sample lead time and production lead
time

Figure 2. AHP hierarchy for determining attribute weights

3.2.2. Determining relative priorities and weights

The purpose of attribute weights is to express the importance of each attribute relative to others in retrieving
the most appropriate case. It is difficult for the decision maker to assign relative weights for all attributes
simultaneously, which can be overcome by pair-wise comparisons in AHP. It involves construction of a
square matrix in which the set of attributes are compared pair-wise and the weights w1, w2,…., wn associated
with each attribute are calculated by geometric mean.

The sum of weights of individual attributes in a particular group is normalised to one. The sum of weights
of all attribute groups is also normalised to one. The overall weight of any particular attribute is equal to the
product of its own weight and the weight of the attribute group it belongs to.

4
3.2.3. Checking the consistency of pair wise comparison

The consistency ratio is calculated using the maximum eigen value, λmax (please refer the appendix). As
suggested by Saaty (1994), the consistency ratio can be tolerated if its value is below 0.10, otherwise it
would bias the result by considerable margin, and the comparison would need to be redone carefully.

3.3. Specifying attribute values

The values of quantitative attributes (such as casting size, casting weight and minimum section thickness)
are specified as product requirements by the design engineer. However, shape complexity that plays an
important role in process planning is difficult to express in numbers. In fact, there appears to be no
quantifiable definition of shape complexity in technical literature. An interesting part of our work is
expressing the same on a 0-100 scale in which lower values stand for simple shapes while higher values
represent complex shapes.

3.3.1. Part shape complexity

This is estimated from the 3D model of the casting. For this purpose, an equation to calculate shape
complexity in terms of surface area and number of cored features has been developed using regression
analysis. To eliminate the effect of the absolute value of surface area (for example a cube of any size will
have the same shape complexity), it is expressed in terms of area ratio. The proposed equation is given by:
Y = α o + α1 × Ca + α 2 × Cc (2)
where,
Y = Complexity index
αi = Regression coefficients
Ca = Area ratio
= 100 x (1 – ( surface area of cube of equal volume / surface area of solid))
Cc = Core complexity factor = 100 × ⎡1 − 1
⎣ ( )
1 + n ⎤ , where n = number of cores

The procedure adapted for developing equation is explained below:

(a) A set of components involving simple to complex shapes were selected.

(b) Relative shape complexity was determined for each component (by pair wise comparison using AHP)
by CAD engineers and tooling experts, and average values were taken. Some of these models and their
shape complexity values are shown in figure 3.

(c) The value of area ratio (Ca) for each component was determined from the solid model. The core
complexity factor (Cc) was determined based on the number of cores. The corresponding values for
different parts are shown in table 1.

(d) From the above values, an equation for shape complexity was developed as:
Y = 0.3 × Ca + 0.8 × Cc − 14 (3)
For very simple castings without any cored features, the equation may give shape complexity value as
below zero; in such cases, the value is set as zero. For all other shapes, the equation was found to give
satisfactory results.

(e) For determining the shape complexity for a new component, values of Ca and Cc are determined from
its 3D model and then shape complexity is calculated using the above equation. For example, for yoke
casting (figure 4) the surface area, volume of casting and number of cored features are determined as
41571.1 mm2, 209083 mm3, and 3 respectively. The surface area of a cube with the same volume is
21136.4 giving an area ratio of 49. Finally, equation 3 gives the shape complexity of yoke as 41. The
perceived (expert opinion using AHP) and predicted shape complexity (using the proposed equation)
are shown in table 1.

5
(a) (b) (c) (d)

Figure 3. Shape complexities of sample components


(a) Rim - 16, (b) Lug - 27.5, (c) Differential casing - 55, (d) Steam valve - 78

Table 1. Shape complexity

Part Part Part Surface Area Nu Core Calculated Perceived


name volume surface area of ratio m complexity shape shape
(mm3) area cube with (Ca) of factor complexit complexity
(mm2) same core (Cc) y by AHP
volume s
(mm2)

Cube 1000000 60000 60000 0 0 0 0 6.0


Plate 200000 46000 20520 55.39 0 0 3.0 8.0
Sphere 523333 125600 38964 44.00 0 0 0 10.0
Cube with hole 360000 72800 30364 58.29 1 29.29 27.0 12.0
64632645 1074742 4485242
Rim 58.27 1 29.29 27.0 16.0
4 8 2
Step bearing 317958 64473 27951 56.65 1 29.29 26.0 17.5
Pulley 4832011 373105 171489 54.04 1 29.29 27.0 20.0
Body cap 1845480 233402 90273 61.32 1 29.29 28.0 24.0
Stand 870364 92587 54695 40.93 2 42.26 32.0 25.0
Bracket 930670 100397 57194 43.03 2 42.26 32.0 25.0
Lug 322676 45168 28227 37.51 3 50.00 37.0 27.5
Ball valve 980197 149210 59205 60.32 2 42.26 38.0 32.5
Globe valve 10102452 1352865 280394 79.27 3 50.00 50.0 35.0
Auto cylinder 613155 200916 43304 78.45 1 29.29 33.0 42.0
Knuckle 102494 110749 13141 88.13 2 42.26 47.0 48.0
Differential
558928 172788 40712 76.44 2 42.26 43.0 55.0
casing
Unloader valve 167186 45622 18209 60.09 6 62.20 54.0 60.0
End shield 3713 67291 1439 97.86 4 55.28 60.0 63.5
Steam valve 38047221 2430200 678734 72.07 10 69.85 64.0 78.0
Engine block 46936080 4932554 780709 84.17 25 80.39 76.0 92.0

6
Figure 4. Yoke casting
3.4. Case retrieval

For faster retrieval, the case base is first screened for compatible projects, followed by identifying the
nearest neighbour.

3.4.1. Screening the process data base and case base

The case base is screened to determine a set of feasible cases that have the same material-process
compatibility with that of the new project. For this purpose, the feasible processes for the new project are
first identified by comparing the attribute values (such as casting material, weight, minimum section
thickness, surface finish, tolerance and delivery quantity) of the new project with the corresponding
characteristics of different metal-specific processes stored in the process database. The casting processes
presently covered in the database include green sand casting, high pressure sand casting, shell molding, no
bake process, gravity die casting, low pressure die casting, high pressure die casting, wax investment
casting and foam investment casting. The materials include gray iron, ductile iron, steel, alloy steel,
aluminum, copper, magnesium and zinc alloys.

The case base (database of old projects) is then screened to shortlist the cases that are manufactured by any
one of feasible processes (for the new case). Only short-listed cases are considered for the next step, that is,
the nearest neighbour identification. This helps in reducing the overall time for retrieving the most
appropriate case, and prevents the accidental retrieval of a case that may have the same overall index
(explained later) but an incompatible process for the new case.

3.4.2. Determining the nearest neighbour

Weights are calculated for the above attributes (except casting material) using the analytical hierarchy
process as explained earlier. After specifying the attribute values, the similarity of the new casting to a
casting in the case base is determined using Euclidean distance function as follows:
n
Sim( N , O ) = ∑ wi × sim( ni , oi ) (4)
i =1

sim ( ni , oi ) = 1 − dist ( ni , oi ) (5)

( ni − oi )
2
dist ( ni , oi ) = (6)
where,
Sim(N,O)= Similarity of new case to old case in the case base
sim(ni, oi)= Similarity of individual attribute values of new case to old case
dist(ni, oi)= Distance between individual attribute values of new case to old case
ni = Value of attribute i of old case in case base
oi = Value of attribute i of new case

7
wi = Weight of attribute i
n = Number of attributes

In a similar manner, the similarity values are determined for all short-listed cases (casting projects) using
the weights and the corresponding attribute values. Based on these similarity values, the nearest three cases
(with respect to the new case) are identified. The user can select any one of the nearest neighbour cases
displayed by the process planning system. For increasing the accuracy of the nearest neighbour retrieval, the
similarity at the attribute level is also checked and only the cases having similarity of 75% or more at
attribute level are considered for retrieval.

3.4.3. Normalising the attribute values

As evident from the above equations, attributes with very small values (like tolerance and surface finish)
exert very little influence on the similarity coefficient compared to attributes with large values (like casting
weight, casting size and order quantity). But they do play an important role in process planning and cannot
be ignored in finding the nearest neighbour. To overcome this problem, all attribute values are normalised
to a scale 0 to 1 by designing appropriate functions.

Figure 5. Mapping attribute values

Table 2. Scale for mapping attribute values

Attribute name Value range and unit


Casting size 1-5000 mm
Casting weight 0.10-10 000 kg
Minimum section thickness 0.50-100 mm
Maximum section thickness 10-500 mm
Shape complexity 1-100
Minimum core size 1-100 mm
Maximum core size 1-1000 mm
Surface roughness 1-100 µm
Maximum void size 0.01-10 mm
Tolerance 0.1-10 mm per 100 mm
Order quantity 1-1 000 000
Production rate 1-100 per hour
Sample lead time 1-100 days
Production lead time 1-100 days

8
If Amin, Amax and Aact are the minimum, maximum and actual values respectively of attribute Ai (figure 5),
then the mapped value defined by function f(Ai) is given by:

1..............................................if Aact ≥ Amax


f ( Ai ) = ( Aact − Amin ) ( Amax − Amin )...........if Amax ≥ Aact ≥ Amin (7)
0...............................................if Aact ≤ Amin

The extreme values Amax and Amin of various attributes ascertained from various sources are given in table 2
(Bralla 1988, Heine and Rosenthal 1991). The values are selected such that most castings will fall within
the range.

3.5. Case adaptation and case retention

The next step is to adapt the process plan of the previous case to the new project at hand. If the new project
is similar to a previous one, then the old process plan can be used directly and no adaptation is needed. Case
adaptation involves modifying a process plan using the process planning library. The library stores
information related to steps and process parameters required to perform each casting activity (such as core
making, sand preparation and melting) using different methods. For example, a library corresponding to
core-making contains process planning information related to different core-making methods (such as CO2
method, oil sand, hot-box and cold-box). A particular option from the library is selected interactively and
copied to the existing project to modify its process plan. After case adaptation and modification, the new
case is added to the case base for future reference. Thus, new experience is continuously added to the case
base, making the system gradually more powerful.

4. IMPLEMENTATION AND RESULTS

The overall methodology has been implemented in a web-based framework called WebICE (Web-based
Integrated Casting Engineering) developed in our laboratory (Akarte and Ravi 2002). The WebICE
facilitates web-based creation, updating and exchange of casting project data. The project data is stored in a
Casting Data Markup Language (CDML), defined using the XML standard (Ravi and Akarte 2002). The
CDML consists of two parts: CDML tree and data blocks. The CDML tree represents the hierarchical
relationship between different types of information essential for collaboration between product, tooling and
foundry engineer, whereas the data blocks are used for storing the actual project data. The hierarchical tree
structure enables easy identification of the desired information. A library function offers alternative options
for the set of field values in a data block. The library database contains several data block files (options)
corresponding to a particular CDML data block. Any of these can be selected and its values can be copied
to the corresponding data block in the project. This eliminates manual input and thereby the possibility of
human errors.

The case-base is composed of cases, which are in a predefined structure of WebICE framework. A case is
composed of two major parts: problem description and solution. The problem description refers to attributes
in the case that are required to describe the problem. These attributes and their corresponding values are
stored under PRODUCT node of CDML tree. The solution refers to the process plan and is stored under
PROCESS node.

As mentioned earlier, 15 attributes have been identified for the casting process planning and all the
attributes (except cast material) are divided into three groups namely, geometry related, quality related and
production related attributes. Geometry related attributes (maximum casting size, casting weight, minimum
and maximum section thickness, minimum and maximum core size, shape complexity) have been computed
automatically from a 3D model of the casting (created in a standard CAD system and saved in STL format).
Production and quality related attribute values have been specified interactively by the user. The values of
these attributes for a body cap casting (figure 6) are shown in table 3.

9
Figure 6. Body cap casting

Table 3. Input for process planning

Attribute Value
Material Grey cast iron
Maximum casting size 255 mm
Weight 14.35 kg
Minimum section thickness 12 mm
Maximum section thickness 36 mm
Minimum core size 46 mm
Maximum core size 95 mm
Shape complexity 28
Maximum void size 0.5 mm
Surface finish 8 μm
Tolerance 1 mm
Order quantity 5000 units
Production rate 25 per hour
Sample lead time 45 days
Production lead time 15 days

10
Figure 7. Weight calculation using AHP

As mentioned earlier, weights of these attributes were calculated using AHP. The pair wise comparison for
calculating the relative weights for quality related attributes is shown in figure 7. For example, ‘maximum
void size’ was given importance of 2 when compared with ‘dimensional tolerance’ (note that 1 means equal
importance, and 9 means extremely important compared to the other). Similarly, ‘maximum void size’ is
given moderate importance of 3 over ‘surface roughness’. Both ‘Dimensional tolerance’ and ‘surface
roughness’ are assigned equal importance value 1. After the pair wise comparison, the consistency ratio and
the relative weight of each attribute are calculated (please refer the Appendix). In this case, the consistency
ratio is 0.015, which implies slight inconsistency in pair wise comparisons, though well within the
acceptable limit of 0.10. After calculating the geometric mean and normalising, the relative weight of
‘maximum void size’ was 55% followed by ‘dimensional tolerance’ (24%) and ‘surface roughness’ (21%).
Similarly, relative weights of attributes belonging to other groups (geometry and production) as well as
relative weights of the groups are determined. The weight of an attribute is given by the product of the
relative weight of each attribute and the relative weight of the group to which it belongs. The attribute
weights are summarised in table 4.
Table 4. Weights of attributes
Group Relative Attribute Relative Attribute
Attributes weight Weight weight

Casting weight 10.62 5


Maximum size 9.46 5
Min. section thickness 26.71 13
Geometric Max. section thickness 14.34 7
50.00
attributes
Min. core size 15.02 8
Max. core size 8.84 4
Shape complexity 15.02 8

Dimensional tolerance 24.02 6


Quality attributes 25.00 Surface roughness 20.98 5
Maximum void size 54.99 14

Order Quantity 45.31 11


Production Production rate 26.16 7
25.00
attributes Sample lead time 16.71 4
Production lead time 11.82 3

11
After submitting the attribute values, the CBR methodology is executed to determine the feasible processes
and the nearest neighbours. If the nearest neighbours are found suitable, the user can select one of them to
copy its process plan to the new project without adaptation. The final results of case based reasoning
algorithm are shown in figure 8. In the present example, only one nearest case has been selected.

Figure 8. Nearest neighbour algorithm result

If the nearest neighbour is not found suitable for direct application to the new case, then some modifications
are required. As mentioned earlier, it requires a process planning library for each activity. Such a library for
core making activity is shown in figure 9. The user can select a particular option from the library and copy
it into the current project. Finally, the process plan along with the product details is added and retained in
the case base, and can be used for future reference and retrieval.

Figure 9. Library for core making

12
5. CONCLUSIONS AND FUTURE WORK

Our work was motivated by the need to develop a systematic approach to casting process planning in a 3D
design environment, which has received very little attention so far. The case based reasoning methodology
in conjunction with the nearest neighbour algorithm and analytical hierarchy process (for assigning weights
to attributes) proved to be effective and efficient. The computation time was further reduced by matching
the casting process of previous cases with the feasible process of the new case. An equation has been
developed to quantify the casting shape complexity that plays an important role in process planning. The
equation has been tested for number of castings and found to give results comparable to an expert’s
judgement. The primary input being the product design data, the system can be used by design engineers for
manufacturability analysis through early cost estimation and quality prediction by simulation, both of which
require process planning information. Further, the web based implementation of the methodology enables
collaborative engineering between product and process engineers irrespective of their physical location. The
future direction of the work involves making the process planning system more intelligent to minimise
human interaction. For this purpose, a knowledge management system is being developed, which will
provide facilities to add domain knowledge in the form of equations and if-then rules. These rules will be
used to semi-automatically adapt a retrieved case by selecting the most appropriate options from the library
of process plan steps and parameters. To the best of our knowledge, the work reported in this paper is the
first of its kind, and we hope that it will spur further research in this area by both casting and CAD/CAM
communities.

REFERENCES

Aamodt, A., Plaza, E., 1994, Case based reasoning: Foundational issues, methodological variations and
system approaches. AI Communications, 7 (1), 39-59.
Ajmal, A., Dale, B., 1987, A computer-aided foundry specific planning and estimating system: Its
development and some operating results. Engineering Costs and Production Economics, 13 (1) 55-63.
Akarte, M., Ravi, B., Creese, R., 1999, Casting process selection using AHP and fuzzy logic. International
Seminar on Manufacturing Technology Beyond 2000, Bangalore, India, November 18, 1999.
Akarte, M. M., Ravi, B., 2002, Web-based collaborative engineering of cast products. 30th Computers and
Industrial Engineering Conference. Tinos Island, Greece, June 28, 2002.
Bergmann, R., Breen, S., Goker, M., Manago, M., Wess, S., 1998, Developing Industrial Case Based
Reasoning Applications- The INRECA Methodology (Berlin: Springer).
Bralla, J., 1988, Handbook of product design for manufacturing (New York: McGraw-Hill)
Chang, T., Wysk, R., 1985, An Introduction to Computer Aided Process Planning Systems (NJ: Prentice
Hall).
Chang, H., Dong, L., Liu, F., Lu, W., 2000, Indexing and retrieval in machining process planning using
case based reasoning. Artificial Intelligence in Engineering, 14, 1-13.
Chougule, R.G., Ravi, B., 2004a, Collaborative Design for Manufacture – Metal casting applications.
Proceedings of IEEE International Conference on Intelligent Engineering Systems, Cluj Napoca,
Romania, September 19-21.
Chougule, R.G., Ravi, B., 2004b, Casting cost estimation in an integrated product and process design
environment. Communicated to International Journal of Computer Integrated Manufacturing.
Creese, R. C., 1979, Group technology for higher productivity and cost reduction in the foundry. AFS
Transactions, 87, 227-230.
Darwish, S. M., El-Tamimi, A. M., 1996, The selection of casting process using an expert system.
Computers in Industry, 30, 77-86.
Department of Energy (DoE), 1999, Energy and Environmental Profile of US Metal Casting Industry, A
report published by Office of Industrial Technology, Department of Energy, USA.
ElMarghy. H. A., 1993, Evolution and future perspectives of CAPP. Annals of the CIRP, 42(2), 739-751.
Er, A., Sweeney, E. T., Kondic, 1996, A knowledge based system for casting process selection. AFS
Transactions, 104, 363-370.
Gu, P., Norrie, D.H., 1995, Intelligent Manufacturing Planning (London: Chapman & Hall).
Heine, R.W., Rosenthal, P.C., 1991, Principles of Metal Casting (New York: McGraw-Hill).
Kim, K., Han, I., 2001, The cluster indexing method for case based reasoning using self organizing maps
and learning vector quantization for bond rating cases. Expert Systems With Applications, 21, 147-156.
Kolodner, J. L., 1991, Improving human decision making through case based decision aiding. AI Magazine,
12 (2), 52-68.
Law, T.D., et al, 1981, Third report of working group (E4): Designing and using casting classification
system. British Foundryman, 74, 145-152.

13
Lei, Y., Peng, Y., Ruan, X., 2001, Applying case based reasoning to cold forging process planning. Journal
of Material Processing Technology, 112, 12-18.
Marefat, M., Britanik, 1997, Case based process planning using an object-oriented model representation.
Robotics and Computer-Integrated Manufacturing, 13 (3), 229-251.
Markus, A., Vancza, J., Horvath, M., 1997, Process planning by retrieval and adaptation, Computers in
Industry, 33, 47-60.
Optiz, H., 1970, A classification system to describe work-pieces. (Oxford: Pergamon Press).
Ravi, B., Akarte, M. M., 2002, Casting Data Markup Language for web-based collaborative engineering.
AFS Transactions, 110, 93-108.
Saaty T. L., 1994, How to make decisions: The analytic hierarchy process. Interfaces, 24 (6), 19-43.
Sirilertworakul, N., Webster, P., Dean, T., 1993, A Knowledge base for alloy and process selection for
casting. International Journal of Machine Tools Manufacturing, 33 (3), 401-415.
Tiwari, M., Rama, K., Bhatnagar, S., 2001, A case based computer-aided process-planning system for
machining prismatic components. International Journal of Advanced Manufacturing Technology, 17
(6), 400-411.
Wettschereck, D., Aha, D., Mohri, T., 1995, A review and empirical evaluation of feature weighing
methods for a class of lazy learning algorithms. Artificial Intelligence Review, 11 (15), 273-314.
Yaskovskii, I. G., Krivitskii, V. S., Vasil’ev, A. A., 1971, Technical and economic principles of improving
variable batch casting production, Russian Castings Production, 10, 409-410.
Younis, M. A., Waheb, M.A., 1997, A CAPP system for the rotational component. Computers and
Industrial Engineering, 33, 509-512.
Zhang H., Alting L., 1994, Computerised Manufacturing Process Planning Systems (London: Chapman &
Hall).

APPENDIX: Analytical Hierarchy Process

The steps to find the vector of weights and check the consistency of judgment (Saaty, 1994) are as follows:

1. Construct a pair-wise comparison matrix: Assuming n criteria, the pair-wise comparison of criterion i
with criterion j yields a square matrix of criteria called A1nxn. Where, aij is the element in the pair-wise
comparison matrix, giving comparative importance of criterion i with respect to criterion j (table 5). In
matrix A1, aij = 1, when i = j and aji = 1/ aij .
⎡ a11 a12 ........... a1n ⎤
⎢a a 22 ........... a 2 n ⎥⎥
A1n×n = ⎢ 21
⎢..... ..... ........... ..... ⎥
⎢ ⎥
⎣a n1 a n 2 ........... a nn ⎦

[Insert table 5 here]

2. Find the weight of each criterion (Wj) by calculating geometric mean of ith row (GMi) and then obtain the
relative weights of each criterion by normalising geometric means of rows in the comparative matrix.
1

⎡n ⎤
n
n

GM = ∏ aij

i

and W = GM
i i ∑ GM i

⎣ j =1 ⎦
i =1

3. Calculate matrices A3 and A4, such that A3 = A1 × A2 and A4 = A3 A2 ,

[ w1 w2 .......... wn ]
T
Where, A2 =
4. Find out the maximum eigen value (λmax ) , which is the average of matrix A4.
5. Calculate the consistency index (CI ) = (λ max − 1) (n − 1)
6. Obtain the random index (RI) for the number of criteria used in decision-making from the table 6.
[Insert table 6 here]

7. Finally, calculate the consistency ratio (CR ) = CI RI . Usually, a CR of 0.10 or less is considered
acceptable.

14

You might also like