Product Design for the Environment: A Life Cycle Approach - Chapter 14 ppt

375
Chapter 14
Optimal Disassembly Planning
From the viewpoint of environmental protection, an effi cient planning of
disassembly operations takes on strategic importance, since it can improve
both the useful life of the product (facilitating service interventions) and the
end-of-life phase (favoring the recycling of materials and the reuse of compo-
nents). After an overview of the main issues and current approaches to the
problem of Disassembly Planning, this chapter proposes a new approach
aiding the identifi cation of optimal disassembly in relation to service opera-
tions and the recovery of resources at the end-of-life. Despite their diverse
aims, the two tools proposed here have in common both the modeling
typology on which they operate and the logic underlying the search algo-
rithms used. The choice of genetic algorithms is dictated by the complexity
inherent in the complete mathematical solution of the problem of generating
the disassembly sequences, which suggests the use of a nonexhaustive
approach.
The results of a wide-ranging series of simulations indicate that both tools
can be used not only for the characteristic purposes of disassembly planning,
but also for aiding design. This is particularly true for the second tool, which
offers a complete approach to the problem of disassembly aimed at recovery;
it combines evaluations of economic and environmental impacts and extends
these evaluations over the product’s entire life cycle. Its structure provides
the designer with an autonomous capacity to decide on both the level of
disassembly to be achieved and the defi nition of the optimal recovery plan
(i.e., the best destination for the disassembled components on the basis of
their properties).
The fundamental issues in this chapter were previously introduced in
14.1 Disassembly Planning
The disassembly of products is necessary every time it is appropriate to
remove subsystems or single components comprising the product itself. From

2722_C014_r02.indd 3752722_C014_r02.indd 375 11/30/2005 1:52:47 PM11/30/2005 1:52:47 PM
© 2006 by Taylor & Francis Group, LLC
Chapters 9 and 13.
the viewpoint of environmental protection, disassembly can have several
goals (Lambert, 1997): recovering parts, components, and subassemblies that
can be reused in new products; recovering recyclable materials; and accessing
parts or components that may be subject to service operations (repair,
maintenance, diagnostics).
As highlighted in the previous chapter, interventions to improve the
process of product disassembly can be made at two different levels (Jovane
et al., 1993; Gungor and Gupta, 1999):
• At the design phase, adopting choices that can favor the disassembly
of the constructional system (Design for Disassembly—DFD)
• Seeking to best plan and optimize the process of disassembly
(Disassembly Process Planning—DPP)
Disassembly Process Planning, which includes all the problems relating to
disassembly of constructional systems, is considered in terms of two differ-
ent levels of analysis (Lambert 2003):
• Sequence Level—Starting from a mathematical representation of the
assembled system, the analysis considers the problem of generating
and optimizing the sequence of disassembly (Disassembly
Sequencing).
• Detailed Level—Starting from the physical and geometric properties
of components and fasteners, disassembly analysis takes into consid-
eration the handling of components, directions of disassembly,
conditions of obstruction, and choice of equipment, as far as deter-
mining the trajectories of any possible automatic manipulators
(Disassembly Path Planning).
In a more complete view, these aspects should also be complemented by an
analysis of the optimal disassembly level (Disassembly Leveling), which is

the study of the level of disassembly that bests reconciles the requirements
and the advantages of disassembling a system’s components with the costs
Section 13.1. The literature contains general evaluations of cost–revenue
curves for disassembly interventions aimed at repair and recovery at end-of-
disassembly become prohibitive as the level of disassembly increases. On the
contrary, beyond a certain level the revenues tend to stabilize. Consequently,
the profi t curve has a maximum, after which it tends to decrease as the level
of disassembly increases.
The profi t of recovery can be improved if the product architecture is such
that the fi rst disassembly results in freeing the most critical or most valuable
376 Product Design for the Environment
2722_C014_r02.indd 3762722_C014_r02.indd 376 11/30/2005 1:52:47 PM11/30/2005 1:52:47 PM
© 2006 by Taylor & Francis Group, LLC
entailed by this disassembly. This concept was introduced in Chapter 13,
life (refer to Chapter 9, Figure 9.5). According to these evaluations, the costs of
Optimal Disassembly Planning 377
parts. This consideration highlights the direct dependence of the effi ciency
of disassembly on the product architecture, and particularly on the depth
of disassembly of single components in relation to certain characteristics
that can make their removal opportune or necessary during use or at end-
The profi t function of disassembly is also strongly sensitive to the effi ciency
of the disassembly sequence. With a judiciously optimized disassembly
sequence it is possible to:
• Reduce the times and costs of disassembly operations
• Perform the most appropriate disassemblies, avoiding those that are
superfl uous or less profi table
This is, therefore, the fi rst intervention factor for profi table disassembly. It is
indirectly conditioned by the characteristics of the product’s architecture,
which, depending on the typology, can favor the effi ciency of the disassembly
sequence.

In conclusion, the problem of disassembly planning includes the main
aspects:
• Analysis of the system characteristics (component geometries, rela-
tions between components, junctions)
• Generation of possible disassembly sequences
• Identifi cation of the most effi cient and economical disassembly
sequences
• Determination of optimal level of disassembly
14.1.1 General View of the State of the Art
Early approaches to the problem of disassembly planning were developed
on the basis of understanding previously acquired in relation to the prob-
lems associated with assembly. These approaches came from the assump-
tion that disassembly sequences could be assimilated to assembly
sequences in reverse (Homem De Mello and Sanderson, 1991). However,
various authors have subsequently described the profound differences
existing between the two problems (principally, the fact that assembly
processes are often not completely reversible, and the frequent need for
selective or partial disassembly). These two problems demonstrate the
need to treat disassembly planning in a specifi c and more appropriate
manner (Srinivasan et al., 1997; Dini et al., 2001; Lambert, 2003). With this
new perspective, a large number of approaches to disassembly have been
proposed over the last decade, suggesting diverse attempts at classifi cation
2722_C014_r02.indd 3772722_C014_r02.indd 377 11/30/2005 1:52:48 PM11/30/2005 1:52:48 PM
© 2006 by Taylor & Francis Group, LLC
of-life. Depth of disassembly was thoroughly discussed in Chapter 13.
(O’Shea et al., 1998; Tang et al., 2000; Lambert, 2003). In particular, authors
have emphasized:
• The different levels of approach (oriented toward components, or
oriented toward products)
• The differences in the aims of disassembly (maintenance and

servicing, or removal and recovery at end-of-life)
• The differences in disassembly modeling, in generating the disas-
sembly sequences, and in methods to identify the optimum
solution
14.1.2 Extension to Design of the Life Cycle
In environmental terms and in relation to the phases of a product’s life cycle,
the great importance of disassembly is its functionality in the phases of use
(supporting maintenance and servicing operations) and end-of-life (support-
ing operations of recovery and disposal). With regard to servicing operations,
several different approaches are described in the literature (Subramani and
Dewhurst, 1991; Yokota and Brough, 1992; Vujosevic et al., 1995). While these
have in common the criterion of determining the optimum disassembly
sequence based on the minimization of costs, they differ in the typologies of
models used to represent the disassembly process (connection diagrams or
component–junction diagrams, “AND/OR” graphs). However, the problem
is generally rooted in planning the selective disassembly of components
requiring maintenance.
The subsequent consideration of the operations of recovery and disposal at
end-of-life required an extension of the problem of disassembly planning.
Here, disassembly sequences must be optimal not only in strictly economic
terms, but also in terms of environmental impact of the end-of-life phase
(Gungor and Gupta, 1999). This led to the introduction of the concept of
Recovery Planning, based on a quantitative evaluation of a product’s value
at end-of-life in terms of its potential for reuse, remanufacturing, and the
recycling of materials (Navin-Chandra, 1993; Navin-Chandra, 1994; Zussman
et al., 1994; Pnueli and Zussman, 1997). These approaches all use graphic
models to represent disassembly processes.
In contrast, other approaches tend toward analytical mathematical
modeling. Beginning with component–junction diagrams, these introduce
matrix representations to express the precedences of component disassem-

bly (Disassembly Precedence Matrices) (Zhang and Kuo, 1997). This allows
the elaboration of more data and makes it possible to study the problem at
a greater level of detail, going so far as to consider even the times and
costs of changing a tool or the direction of disassembly. Nearly optimal
378 Product Design for the Environment
2722_C014_r02.indd 3782722_C014_r02.indd 378 11/30/2005 1:52:48 PM11/30/2005 1:52:48 PM
© 2006 by Taylor & Francis Group, LLC
Optimal Disassembly Planning 379
disassembly sequences can then be developed on the basis of a much more
detailed analysis than those possible with graphic models (Gungor and
Gupta, 1998). The introduction of matrix modeling also allows direct inte-
gration with CAD modeling of the assembly under examination (Dini
et al., 2001).
14.1.3 Application of Artifi cial Intelligence
Extending the problem of disassembly planning and placing it in the context
of the product’s life cycle results in a signifi cant increase in the amount of
data to be elaborated and in the complexity of the performances to be opti-
mized. To deal with the resulting increased complexity of the problem,
researchers have turned to certain instruments of artifi cial intelligence
(Lambert, 2003): neural networks, fuzzy logic, and genetic algorithms. The
latter, in particular, have been applied in various cases, by virtue of their
characteristic of working effi ciently in open research domains and of identi-
fying solutions close to optimal. In this context, some studies have described
the use of genetic algorithms as aids in:
• The optimization of disassembling components or subassemblies for
the purpose of maintenance, based on a component–contact constraints
diagram model (Li et al., 2002)
• The economic and environmental analysis of disassembly processes,
based on an AND/OR graph model and converting environmental
factors into economic costs (Seo et al., 2001)

• The defi nition of the most effi cient disassembly strategy in relation
to the requirements of recovery at end-of-life, based on criteria of
maximum revenue and minimum number of components to be
disposed of as waste (Caccia and Pozzetti, 2000)
14.1.4 Concluding Considerations
The great variety of methods proposed in the literature, summarized above,
is directly correlated to the specifi c problems the various authors intended to
resolve. At present, there is little effort directed at the defi nition of a system-
atic methodology that is able to integrate the different approaches and to
guide in solving different problems using the most effective approach. Apart
from the specifi city of the various methods found over the entire spectrum of
studies on disassembly planning, it is also worth noting the limited vision of
the environmental problem, which in this context is usually treated by trans-
lating environmental aspects into economic costs and considering only the
end-of-life phase in the analysis.
2722_C014_r02.indd 3792722_C014_r02.indd 379 11/30/2005 1:52:48 PM11/30/2005 1:52:48 PM
© 2006 by Taylor & Francis Group, LLC
380 Product Design for the Environment
14.2 Objectives and Approach to the Problem
With clear reference to these observations on the state of the art, this chapter
proposes an approach to disassembly planning characterized by:
• Matrix-type modeling, based on the analysis of component geome-
tries and on the relations of junctions and movement, in a way that
facilitates direct interfacing with conventional CAD modeling of
assemblies
• A calculation program with a structure allowing the identifi cation of
the disassembly sequence that is optimal in terms of the aspects
considered most signifi cant—servicing operations and planning
recovery at end-of-life
The two different aspects treated here, service and the recovery of resources

at end-of-life, require that a distinction be made between the two different
typologies of disassembly (Srinivasan et al., 1997):
• Selective disassembly, where the objective is the disassembly of one
or more preselected components (an approach oriented toward
servicing operations)
• Partial or complete disassembly, where the components to be disas-
sembled are not chosen a priori, but are defi ned by the research algo-
rithm itself on the basis of certain important properties characterizing
the components (an approach oriented toward recovery operations)
With these objectives, it is possible to develop two algorithms to solve the
problem of optimizing disassembly in the two distinct cases. Despite their
different aims, the two proposed tools operate on the same typology of
modeling and share the logic followed in developing the two algorithms. It
is interesting to note, however, that while the fi rst is limited to a conventional
approach to the problem (selective disassembly optimized by minimizing
disassembly times), the second tool deals with the environmental problem of
recovery through an innovative approach, in that it is:
• Based on a complete analysis (bringing together functions of cost
and environmental impact) and extends the evaluations to cover the
entire life cycle
• Characterized by an autonomous capacity to determine both the
level of disassembly to be achieved and the optimal recovery plan
(i.e., the best destination for the disassembled components, on the
basis of their properties)
2722_C014_r02.indd 3802722_C014_r02.indd 380 11/30/2005 1:52:48 PM11/30/2005 1:52:48 PM
© 2006 by Taylor & Francis Group, LLC
Optimal Disassembly Planning 381
14.3 Common Structure of the Proposed Tools
In general, the complete mathematical solution to the problem of generating
disassembly sequences requires such a large number of calculations that it

becomes extremely complex (Lambert, 1997; Moyer and Gupta, 1997). While
heuristic methods can be used to manage this problem, they do not ensure the
determination of optimal solutions (Gungor and Gupta, 1997; Kuo et al., 2000).
To reach the objectives that have been set, the tools proposed here are
directed at the defi nition of optimal (or near optimal) disassembly sequences,
using algorithms of the genetic type (GA—Genetic Algorithms) (Holland,
1975; Goldberg, 1989). This choice is principally motivated by two character-
istics of the research space to be explored that make the use of GA advisable
(Mitchell, 1998): the space is vast and is not unimodal. Starting from this core
choice of GA, the two problems discussed above can be treated by developing
two tools that are distinct but nevertheless share the same calculation code
structure. Furthermore, in both cases the code elaborates the same mathemat-
ical model for the description of the geometries of, and relations between, the
various components (expressed through constraint matrices). The code is thus
able to take into account the changes occurring in the system’s structure
during the progressive disassembly of the single parts. Figure 14.1 shows the
scheme common to both the tools proposed. It postulates a preliminary
FIGURE 14.1 Common schematization of tools.
2722_C014_r02.indd 3812722_C014_r02.indd 381 11/30/2005 1:52:48 PM11/30/2005 1:52:48 PM
© 2006 by Taylor & Francis Group, LLC
382 Product Design for the Environment
modeling, which has the aim of interpreting the assemblies under analysis in
mathematical terms, allowing the subsequent elaboration of this information
by the resolving algorithm. The latter, drawing on a set of data (included in
a database) with varying typologies depending on the tool, identifi es the
optimal solution through three main phases:

• Identifi cation of the system, requiring the formalization of the solu-
tion type to investigate and the defi nition of the objective function
• Generation of possible disassembly sequences

• Identifi cation of the optimal solution
The main elements common to both tools for disassembly planning are
described below. In later sections, they will be described in greater detail,
delineating their more specifi c characteristics.
14.3.1 Common Preliminary Modeling
The system to be disassembled is represented here as consisting of a fi nite
number of independent elements that can be removed individually. System
elements are taken to mean:
• Single components linked to the system by reversible fasteners
• Any possible subgroups of components linked together by irrevers-
ible fasteners
• All reversible fasteners and junction systems (screws, clips, snap-
fi ts, etc.)
With this type of approach, it is not possible to consider subgroups of elements
to be treated as a single entity during the disassembly sequence, unless they
are predefi ned on the basis of their homogeneity, the compatibility of their
Considering a generic system consisting of n elements, the properties rela-
tive to each i-th element E
i
are expressed by an index defi ning the typology
of the element. This index allows one component, whose removal does not
require particular operations other than simple translation in space, to be
distinguished from a fastener, whose removal instead requires a specifi c
intervention with a resulting increase in cost and time. Indicating the total
number of different element typologies by n
e
, each index of element typology
e
j
(which can assume integer values in the range [0, n

e
Ϫ1]) is represented
by data quantifying the diffi culty of removing each element of that type. This
property can be quantifi ed using disassembly times (already assessed by
other authors applying general methods for the evaluation of work times
2722_C014_r02.indd 3822722_C014_r02.indd 382 11/30/2005 1:52:48 PM11/30/2005 1:52:48 PM
© 2006 by Taylor & Francis Group, LLC
materials, or other criteria of affi nity between the elements (see Chapter 11).
Optimal Disassembly Planning 383
[Vujosevic et al., 1995; Kroll and Carver, 1999]), or developing specifi c new
approaches (Sodhi et al., 2004). In this way, each element type can be described
by a mean disassembly time and the corresponding term can be normalized
with respect to the simplest intervention, that of horizontal translation.
characterization adopted in the present study, based on the approximate
disassembly times reported in the literature for the most common types of
unfastening operations (Dowie and Kelly, 1994).

The elements defi ned in this way can be subjected to different elementary
operations; these, too, are described by an index expressing the typology of
disassembly operation. Indicating the total number of operation typologies
by n
o
, it is simple to distinguish between operations given that, once the
diversifi cation of the fasteners and other junction systems has been included
in the analysis of the elements, the operations are reduced exclusively to
movements of linear translation. In the three-dimensional case, such transla-
tions occur along the directions X, –X, Y, –Y, Z, and –Z; therefore, n
o
ϭ 6. Each
generic index of operation typology o

k
, which can assume integer values in
the range [0, n
o
Ϫ1], is associated with an execution time for the operation,
ultimately consisting of translation along the k-th direction. Also in this case,
execution times are normalized with respect to the simplest operation of
horizontal linear translation and can be compiled in a table analogous to
Table 14.1. In this way, if it is considered opportune, it is possible to take into
account the potentially greater diffi culty that may be ascribed to operations
of vertical translation, planning for a longer execution time than that for hori-
zontal translation.
In representing the system in function of disassembly planning, the model-
disassembly depth analysis), based on the geometric characteristics of the
system and on the consequent precedences for disassembly among the
elements constituting the assembly. Although this modeling is based on a
geometric-type approach for the analysis of movements (Woo and Dutta,
TABLE 14.1 Indices of element typologies and characterization
ELEMENT TYPOLOGY
INDEX (e
j
) DESCRIPTION MEAN TIME (sec)
NORMALIZED
TIME (tne
j
)
0 Component (to remove)
1.25 1
1 Screw (to remove) 0.6 (per turn) 0.48 (per turn)
2 Snap–fi t (to open) 1.5 1.2

3 Clip (to remove) 1 0.8
4 Connection (to break) 2 1.6
5 Wires (to disconnect) 1.5 1.2
2722_C014_r02.indd 3832722_C014_r02.indd 383 11/30/2005 1:52:49 PM11/30/2005 1:52:49 PM
© 2006 by Taylor & Francis Group, LLC
ing proposed here is of a matrix type (such as that proposed in Chapter 13 for
Table 14.1 shows the indexing of element typologies together with the
384 Product Design for the Environment
1991; Srinivasan and Gadh, 2002), and makes use of preexisting matrix
models (Zhang and Kuo, 1997; Gungor and Gupta, 1998; Dini et al., 2001),
this modeling is particularly simple and comes down to the compilation of
binary constraint matrices, one for each possible direction of disassembly. In
the case of a system consisting of n elements, for the disassembly direction X
this matrix is expressed by:


Vv
X
ij
i 1,2,
,
j 1,2,
,
ϭ
ϭ
ϭ
⎡
⎣
⎤
⎦

…
…


(14.1)

where the term v
X
ij
has a value of 1 if the j-th element obstructs the removal
of the i-th element in the X direction; otherwise it is 0. Analogously, the spatial
constraint matrices can be defi ned in the other directions (in the case of three-
dimensional analysis there will be six matrices, one for each direction of
disassembly X, –X, Y, –Y, Z, and –Z).
Concerning the compilation of these matrices, it should be specifi ed that:
• If the i-th element is a component, it is necessary to consider as
obstacles all the other components impeding its movement in the
direction under examination, and all the fasteners and junction
systems acting directly on it, constraining it to other components
(the latter may also not appear as direct obstacles to the movement
of the component in question).
• If the element is a fastener, it is necessary to consider as obstacles all
the components impeding its accessibility and movement in the
direction under examination.
14.3.2 Disassembly Sequence and Operation Time
Using the preliminary modeling described above, the resolving algorithm
will generate the possible disassembly sequences in a random manner, as
explained in detail below. It will then evaluate the real practicability of each
sequence based on a simple rule: To disassemble any system component, it is
necessary to begin by disassembling the more external components whose

removal is unobstructed, until the fi nal objective is reached. A sequence will,
therefore, consist of a series of element–operation couplets, ordered in such a
way that all the constraint matrices of the type defi ned by Equation (14.1) are
respected. Beginning with the removal of the most external elements, and
updating the constraint matrix step-by-step as the elements are removed,
a correctly ordered (i.e., truly practicable) disassembly sequence is obtained.
On the basis of this statement for the generation of the sequences, it is possi-
ble to defi ne a function that quantifi es the excellence of the sequence. With
this aim, many of the studies present in the literature propose mathematical
2722_C014_r02.indd 3842722_C014_r02.indd 384 11/30/2005 1:52:49 PM11/30/2005 1:52:49 PM
© 2006 by Taylor & Francis Group, LLC
Optimal Disassembly Planning 385
models to evaluate the cost of disassembly, which usually becomes the
objective function to be optimized (or part of it). To quantify the effi ciency of
the disassembly sequences, it is preferable to examine a function that
expresses the total time necessary for the disassembly process, a factor upon
which the cost of disassembly is, in any case, directly dependent (Kroll and
Carver, 1999).
Considering the generic i-th element E
i
, its disassembly time TS
i
is defi ned
by the expression:


TS tne tno cd
ippp
p1
q

i
ϭϩ␦
ϭ
ᠨᠨ
()
∑

(14.2)

where the summation is extended to all the q
i
elements that must be removed
before disassembly E
i
(the total number q
i
includes the element E
i
itself); tne
p

is the normalized disassembly time corresponding to the typology of the p-th
element (this correspondence depends on the index of element typology,
p
to the operation required for the removal of the p-th element (this correspon-
dence depends on the index of operation typology); ␦ is a penalization coef-
fi cient taking into account a possible change in the direction of disassembly
when passing from the removal of element E
p-1
to that of element E

p
(␦ Ͼ 1
expresses the time of changing direction, again normalized with respect to
the operation of horizontal translation); and cd
p
is a binary coeffi cient that
assumes a value of 1 if this change in direction occurs; otherwise it is 0. For a
more thorough analysis, it is possible to introduce additional penalization
coeffi cients that take into account other delaying factors, such as changing
tools or the removal of dangerous or high-value elements requiring greater
care in handling and removal operations (Gungor and Gupta, 1998).
In the context of both tools proposed here, the choice of the disassembly
sequence is based on checking the value assumed by the function (14.2). In
the simplest case of selective disassembly directed at servicing interventions,
function (14.2) becomes the actual objective function of the optimization. In
fact, the problem comes down to the disassembly of a predetermined compo-
nent in the shortest time possible (and thus with the lowest cost). In the case
of disassembly for recovery, function (14.2) will be part of the objective func-
tion, which includes other terms as described below.
14.3.3 Structure and General Characteristics of the Resolving Algorithm
The following description of the resolving algorithm refers to specifi c issues
related to Genetic Algorithm theory. For further details, refer to the special-
2722_C014_r02.indd 3852722_C014_r02.indd 385 11/30/2005 1:52:49 PM11/30/2005 1:52:49 PM
© 2006 by Taylor & Francis Group, LLC
according to Table 14.1); tno is the normalized execution time corresponding
ized resources in the literature (Holland, 1975; Goldberg, 1989; Mitchell, 1998;
Gen and Cheng, 2000).
386 Product Design for the Environment
As stated above, the search algorithm proposed here is of a genetic type
and, therefore, elaborates a set of points (individuals) in the objective func-

tion domain, called the population. The individuals, which represent the
possible solutions, are codifi ed in structures that resemble the confi guration
of a chromosome (Holland, 1975). The structure of the algorithm used in the
present study is entirely conventional and, for clarity, is shown in Figure 14.2,
revealing certain properties that are explained below.
The algorithm initially generates a random population. From this popula-
tion, each execution cycle (generation) selects individuals on the basis of
their fi tness and then applies genetic operators to them, with the most
common being cross-over and mutation. These operators are applied accord-
ing to parameters (called “probabilities of execution”) that remain constant
throughout the whole evolutionary process. The fi tness of the individuals is
quantifi ed by the value assumed by a function of fi tness (the fi tness of an
individual is an expression of the objective function to be optimized). In a
conventional structure, the greater the fi tness of an individual, the higher the
probability that it is selected for reproduction (Goldberg, 1989). Having
generated a new population, the fi tness of the new generation is evaluated,
on the basis of which either a new cycle is executed or the iteration is halted
and the solution to the problem is identifi ed. This depends on whether or not
the condition defi ned by a criterion halting the algorithm (stop criterion) is
achieved.

On the other hand, beginning with a random population (justifi ed by the
opportunity to initially explore as wide a domain of potential solutions as
possible) can limit the effectiveness of the evolutionary process and its capac-
ity to converge toward valid solutions (i.e., those representing disassembly
sequences that are really feasible). This choice is, however, corroborated by
the studies of authors applying genetic algorithms to the optimization of
FIGURE 14.2 Structure of genetic algorithm implemented.
2722_C014_r02.indd 3862722_C014_r02.indd 386 11/30/2005 1:52:49 PM11/30/2005 1:52:49 PM
© 2006 by Taylor & Francis Group, LLC

Optimal Disassembly Planning 387
assembly sequences. In this respect, it should be noted how research has
moved from evolutionary algorithms using already practicable initial
sequences (Bonneville et al., 1995) to those adopting random initial popula-
tions while still obtaining convergence on optimal solutions (Dini
et al., 1999).
In addition to its conventional structure, other particular characteristics of
the genetic algorithm used here are:
• The fi tness of the population is evaluated at each step of the evolu-
tionary process. If the best individual is less fi t than the champion
from the preceding generation, the latter is reinserted in place of
the least fi t individual of the new generation, thereby applying a
technique of elitism in the selection phase (elitist selection)
(De Jong, 1975).
• The initial population is generated in a completely random manner.
In the case where there are no valid individuals in this fi rst genera-
tion (i.e., no feasible disassemblies and no fi tness of an acceptable
level), the search is guided by inserting into the population an indi-
vidual that is already known to have a good level of excellence
(i.e., a “seed” to facilitate the evolution of the population). This oper-
ation of seeding is equivalent to pointing the algorithm toward an
area of the solutions domain that shows itself to be promising in
terms of the search for the optimal (Oman and Cunningham, 2001).
Finally, regarding the stop criterion of the algorithm, it should be noted that
genetic algorithms do not usually show formal properties of convergence
toward the objective function assuming a stable value (and, therefore, toward
the identifi cation of the corresponding optimal solution). For this reason, in
some cases it is necessary to defi ne a heuristic stop criterion. Here, two alter-
natives are considered:
• Following a conventional approach, the investigation halts when a

preset maximum number of generations or level of fi tness is reached
(Hulin, 1997). In this case, the algorithm stops when one of these two
conditions is achieved.
• Following an approach based on requiring the solutions to be stable
over the course of the evolution of successive generations, two
parameters with preset values are introduced, f
1
and f
2
. The algo-
rithm halts when the fi tness of the best individual in the generation
under examination does not differ by more than f
1
% from the mean
of the maximum fi tness values found for the last f
2
generations.
2722_C014_r02.indd 3872722_C014_r02.indd 387 11/30/2005 1:52:49 PM11/30/2005 1:52:49 PM
© 2006 by Taylor & Francis Group, LLC
388 Product Design for the Environment
14.4 Development of the First Tool: Goals of Servicing
To resolve the fi rst type of problem, that of selective disassembly aimed at
interventions of maintenance and servicing during the product’s useful life,
the tool to be applied here is based on the genetic algorithm previously
described, used to identify the best disassembly sequence for a given target
element in an assembly.
This tool is confi ned to identifying a sequence of disassembly operations
that minimize a simple objective function, given that it quantifi es only the
disassembly times. Despite this, its use is not limited exclusively to the plan-
ning and optimization of servicing operations on preexisting products. If

employed as a tool for simulations in the phase of product design and devel-
opment, it can constitute a valid support for a design intervention aimed at
optimizing the system by facilitating the servicing operations on critical
components, in accord with the fundamental principles of Design for
useful for characterizing the diffi culty of disassembling a component, which
is a basic issue in the method of disassembly depth analysis proposed in
14.4.1 Preliminary Modeling
modeling consists of defi ning the n elements making up the system; their
numeration; characterizing them on the basis of their typology; and compil-
ing the constraint matrices of type (14.1), which describe the system mathe-
matically. The elements are characterized through the compilation of a vector
expressing the properties:


Pp
elem i
i1,2, ,n
ϭ
ϭ
[]
…

(14.3)

Given that the typology is the only property of the elements of interest in this
case, p
i
coincides with the typology index corresponding to the i-th element.
The database of the system is limited to collecting the data relative to the
element and operation typologies, as defi ned in Section 14.3.1 (i.e., ultimately,

the values of the removal and handling times relative, respectively, to the
typologies of the elements and of the disassembly operations).
14.4.2 Identifi cation of the System
Proceeding with identifying the system by the search algorithm, the formal-
ization of the solution type to be investigated becomes a vector of integer
2722_C014_r02.indd 3882722_C014_r02.indd 388 11/30/2005 1:52:49 PM11/30/2005 1:52:49 PM
© 2006 by Taylor & Francis Group, LLC
Serviceability (Chapter 9, Section 9.2.1). This tool could also be particularly
With reference to the general schematization in Figure 14.1, preliminary
Chapter 13 (See Section 13.5.2).
Optimal Disassembly Planning 389
numbers, consisting of couplets of indices (Figure 14.3a). The fi rst corresponds
to the index of the element i, following the preliminary numeration. The
second corresponds to the typology index o
k
of the operation required to
remove that element.
The objective function coincides with the function of disassembly time
expressed by function (14.2), calculated for the fi nal element of the disassem-
bly sequence (i.e., the target element).
14.4.3 Generation of Disassembly Sequences and Identifi cation
of Optimal Solution
The generation of the disassembly sequences is intrinsic to the functioning of
the genetic algorithm. Although generated in a completely random manner,
each individual of each generation (represented by the vector in Figure 14.3a)
actually represents a potential disassembly sequence. If this respects the
constraint matrices of type (14.1), and is therefore actually practicable in the
sense discussed in Section 14.3.2, its fi tness is evaluated through the calcula-
tion of the value assumed by the objective function. Otherwise, the sequence
is not feasible and will be treated appropriately in the evolution of subse-

quent generations.
When the algorithm stops, it has found the optimal (or near optimal)
solution, again expressed in the form given in Figure 14.3a, consisting of a
series of element–operation couplets ordered in such a way that the constraint
matrices are respected. These represent the feasible disassembly sequence
that allows the target element to be removed while minimizing the objective
function.
14.5 Development of the Second Tool: Goals of Recovery
The second tool is similar to the fi rst in its general scheme, in its base
modeling, and in its execution procedure. Unlike the fi rst, however, it is
FIGURE 14.3 Formalization of solution type: (a) selective disassembly; (b)
partial or complete disassembly.
2722_C014_r02.indd 3892722_C014_r02.indd 389 11/30/2005 1:52:49 PM11/30/2005 1:52:49 PM
© 2006 by Taylor & Francis Group, LLC
390 Product Design for the Environment
directed at recovery at the product’s end-of-life, seeking a partial or complete
disassembly where the components to be disassembled are not preset but are
defi ned by the algorithm itself on the basis of certain important properties
characterizing them.
In general, the problem of product recovery at end-of-life can be formu-
recovery plan that can effectively balance the costs of the disassembly and
recovery processes and the resulting profi ts, in terms of resources employed
and recovered. In this second case, therefore, the problem is extended to
include not only the optimal disassembly sequence but also the evaluation of
which elements are worth disassembly and what fi nal destination to assign
to them. This requires the introduction of a multiobjective function that takes
into account the times of the disassembly operations together with other
factors, such as the costs of production, recovery, and disposal of compo-
nents, and the indicators that quantify the environmental impact of every
phase of the product’s entire life cycle. The terms of the objective function

thus depend on new factors:
• The properties of the materials and elements constituting the
system
• The level of disassembly
• The fi nal destination of disassembled components (which can be
reused, recycled, or disposed of as waste)
Furthermore, it will be defi ned in such a way that it is possible to treat the
environmental problem of recovery through a complete analysis, combining
functions of cost and environmental impact and extending the evaluation
over the entire life cycle.
By virtue of these characteristics, the proposed new tool can be effectively
employed in planning the end-of-life phase of preexisting products and as an
aid in the product development phase. In fact, it can be used as a tool for the
simulation and optimization of a system, for design specifi cally aimed at
favoring the end-of-life phase of a new product, or in a more attentive design
intervention that concentrates on the product’s overall environmental perfor-
mance over the entire span of its life cycle. Ultimately, this second tool, while
fulfi lling its role as an aid in disassembly planning like the fi rst, is also an
excellent support tool in Design for Disassembly.
14.5.1 Preliminary Modeling
includes defi ning the elements making up the system, numbering them,
characterizing them in terms of their properties, and compiling the constraint
2722_C014_r02.indd 3902722_C014_r02.indd 390 11/30/2005 1:52:50 PM11/30/2005 1:52:50 PM
© 2006 by Taylor & Francis Group, LLC
lated as follows (Chapter 9, Section 9.3): For a given product, determine the
Again referring to the general scheme in Figure 14.1, preliminary modeling
Optimal Disassembly Planning 391
matrices of type (14.1), describing the system mathematically. In this case,
however, the elements are not characterized through the compilation of a
vector, but of a matrix. This is because different properties must be expressed

for each element. Indicating the number of properties to be taken into account
with n
p
, this matrix can be represented by:



Pp
elem ij
i1,2, ,n
j1,2, ,n
p
ϭ
ϭ
ϭ
⎡
⎣
⎤
⎦
…
…

(14.4)

where, generally, the term p
ij
expresses the j-th property of the i-th element.
In this specifi c case, we consider six different properties (n
p
ϭ 6), defi ned as:

i1
• p
i2
, index of the material constituting the i-th element
• p
i3
, weight W of the i-th element
• p
i4
, reusability ⑀ of the i-th element (assumes a value of 1 if the
element can be reused; otherwise it is 0)
• p
i5
, production cost C
prod
of the i-th element
• p
i6
, environmental impact of producing the i-th element
The compilation of the matrix (14.4) requires certain considerations. Above
all, it is necessary to evaluate the weight of each element, the cost of produc-
tion, and the environmental impact associated with its production (the latter
evaluation will be treated below). The reusability of an element depends on
the designed duration of the element’s useful life, as a ratio of the anticipated
working life of the entire product. If the duration is at least twice the working
life of the product, the element is reusable in the same product. In the case
where an element consists of several components characterized by different
durations, the shortest one must be considered. A more detailed discussion
of the reusability of product components and their possible recovery cycles
Finally, it is necessary to introduce an index expressing the material making

up the i-th element. This requires the creation of a table analogous to Table 14.1
for element typology. In this case, however, each material index is associated
characterization.
Having chosen a congruous number of materials, each with an identifying
index, it is possible to associate each material with the following terms that
characterize it in relation to the requirements of the specifi c case:
• The recyclable fraction ␰ (quantifying the recyclability of the
material)
2722_C014_r02.indd 3912722_C014_r02.indd 391 11/30/2005 1:52:50 PM11/30/2005 1:52:50 PM
© 2006 by Taylor & Francis Group, LLC
will be treated in Chapter 15.
• p , typology index corresponding to the i-th element (Table 14.1)
with a set of data, as shown in Table 14.2, which summarizes the complete
392 Product Design for the Environment
• The costs of disposal c
Dsp
and recycling c
Rcl
, and the revenues from
recycling r
Rcl
per unit weight of material (terms necessary for an
economic evaluation)
• The environmental impacts of disposal ei
Dsp
and recycling ei
Rcl
per
unit weight of material (terms necessary for a strictly environmental
evaluation)

Consequently, the database of the system in this case contains different types
of data:
• Those concerning the typologies of elements and operations, as
defi ned in Section 14.3.1 (i.e., the values of removal and handling
times relative, respectively, to the typologies of elements and disas-
sembly operations)
• Those concerning the complete characterization of the materials, as
defi ned in Table 14.2
the more common materials may be obtained from commercially available
databases, such as that of the material selection software CES
®
(Cambridge
Engineering Selector, Granta Design Ltd., Cambridge, UK). The properties of
environmental impact are discussed in detail below. To complete the consid-
erations on preliminary modeling, it is necessary to complement the opera-
tion typologies o
k
considered in Section 14.3.1 with an additional operation
typology, the null operation (with an execution time of zero). The reasons for
this will be explained below.
14.5.2 Advanced Modeling
The studies on disassembly planning found in the literature are generally
characterized by a limited view of the environmental problem, which is
treated by translating environmental aspects into economic costs, with the
analysis confi ned to the end-of-life phase alone. The tool proposed here
TABLE 14.2 Characterization of materials
MATERIAL
INDEX DESCRIPTION RECYCLABILITY
ECONOMIC
TERMS

ENVIRONMENTAL
TERMS
m Material name
ξ
c
Dsp
c
Rcl
r
Rcl
ei
Dsp
ei
Rcl
2722_C014_r02.indd 3922722_C014_r02.indd 392 11/30/2005 1:52:50 PM11/30/2005 1:52:50 PM
© 2006 by Taylor & Francis Group, LLC
As was suggested in Chapter 12, information on the recyclable fractions of
Optimal Disassembly Planning 393
extends this viewpoint, integrating the cost functions with specifi c functions
of environmental impact and including the potential to extend the analysis
over the product’s entire life cycle. This latter aspect, in particular, deserves
more detailed discussion. In the case where it is necessary to disassemble
and recover components from a preexisting product, it is suffi cient that the
economic and environmental analyses are limited to only the end-of-life
phase in order to identify the most effi cient recovery plan.
In the case where the problem of recovery is already being considered in
the product design and development phase, rather than simply predicting
the product’s behavior only in the context of the end-of-life, it is more appro-
priate to have a complete vision of the product’s behavior over the entire life
cycle. It is, therefore, necessary to quantify the economic costs and environ-

mental impacts beginning with the production phases, because it can happen
that a design intervention aimed at improving the possibility of recovering
the product at end-of-life can also lead to an increase in the environmental
impact of production, annulling the environmental improvement obtained
in the recovery phase. With this premise, the tool proposed requires further
modeling described below, which is similar to the modeling introduced in
14.5.2.1 Functions of the Environmental Impact of the Life Cycle
The properties of a material’s environmental impact, expressed by the terms
previously introduced as ei
Dsp
and ei
Rcl
, can be evaluated using the techniques
calculated using SimaPro 5.0
®
software (Pré Consultants BV, Amersfoort, The
Netherlands). With the same instruments, it is possible to evaluate the envi-
ronmental impact associated with the production of each i-th element EI
Prod i
,
introduced in matrix (14.4). This can be expressed as:



EI ei W ei ei
Prod Mat i Prss i Mchg i
ii i i
ϭϩ␮ϩ ␩
ᠨᠨ ᠨ
()

(14.5)

where ei
Mat i
is the eco-indicator per unit weight of the material constituting
the i-th element (weight is expressed by W
i
) and ei
Pcss i
is the eco-indicator of
the primary forming process per unit µ
i
, representing the characteristic
parameter of the process or the quantity of material processed. For a more
accurate evaluation, it is also possible to consider the impact quota due to
secondary machining processes (the term in parentheses, ei
Mchg i
is the eco-
indicator of the machining process per unit of the characteristic process
parameter ␮
i
). These eco-indicators can also be evaluated using the Eco-
indicator 99 method and SimaPro software. If the i-th element consists of
more than one component, its production impact will be the sum of the
impacts of the single components.
2722_C014_r02.indd 3932722_C014_r02.indd 393 11/30/2005 1:52:50 PM11/30/2005 1:52:50 PM
© 2006 by Taylor & Francis Group, LLC
Chapter 12.
of Life Cycle Assessment (Chapter 4) and the Eco-indicator 99 method, and

394 Product Design for the Environment
Again using the same instruments to quantify the environmental impacts,
given a system composed of n elements, the environmental impact of the life
cycle of the i-th element EI
LC i
can be expressed as:



EI EI EI
LC Prod EoL
ii
ϭϩ

(14.6)

where the fi rst term indicates the environmental impact of production,
expressed by Equation (14.5). The second term indicates the environmental
impact at the end-of-life of element EI
EoL i
and can, in turn, be defi ned as:



EI EI ei
Wei
EoL i Prod i Dsp i
iRcli
iii
i

ϭ␣ Ϫ ϩ ␰
ϩ␰
ᠨᠨᠨ
ᠨ
−
()
()
⋅
⎡
⎣
␤ 1
ᠨᠨᠨ
ᠨ
WeiW
ii Dspi
i
⎤
⎦
()
ϩ␥

(14.7)

All the terms in this expression are already defi ned in Section 14.5.1, except
the binary coeffi cients ␣
i
, ␤
i
, and ␥
i

. These depend on the fi nal destination of
the element. In fact, for each element all but one of these coeffi cients is zero,
so that, depending on the case, Equation (14.7) quantifi es the impact of reuse
(fi rst term), of recycling (second term), or of disposal (third term).
It should be noted that:
• The fi rst term, in reality, expresses a recovery of impact, since reus-
ing an element allows the recovery of the environmental impact
associated with its production.
• The second term quantifi es the impact of recycling the recyclable
fraction of material (this impact can also assume a negative value,
that is, it can also be a recovery of impact), and the impact of dispos-
ing of the remaining nonrecyclable fraction.
• The third term quantifi es the impact of disposing of the material of
the whole element.
• In the case where the i-th element consists of more than one compo-
nent (e.g., subgroups of components linked by irreversible fasten-
ers), the three terms must be calculated separately for each component.
The same is true when evaluating the environmental impact of
production EI
Prod i
.
For the whole system of n elements, it is ultimately possible to express the
environmental impact functions of the entire life cycle EI
LC
, or of only the
end-of-life phase EI
EoL
:

EI EI EI EI
LC LC EoL EoL
i1
n
i1
n
ii
ϭϭ
ϭϭ
∑∑

(14.8)

2722_C014_r02.indd 3942722_C014_r02.indd 394 11/30/2005 1:52:50 PM11/30/2005 1:52:50 PM
© 2006 by Taylor & Francis Group, LLC
Optimal Disassembly Planning 395
14.5.2.2 Recovery Planning
From the analytical point of view, the coeffi cients ␣
i
, ␤
i
, and ␥
i
may be consid-
ered coeffi cients of recovery planning and can be expressed as functions of
the reusability of element ⑀
i
, and of a term that, as will become clear below,
constitutes a variable of the optimization process. This latter variable indi-

cates whether it is preferable that the element under examination is recov-
ered or disposed of as waste. Labeling this term with the binary index s
i
,
called the destination index, it will assume a value of 1 or 0 depending on
whether the corresponding i-th element is destined for disposal or recovery.
The recovery planning coeffi cients can then be defi ned as:



␣ϭ␧ Ϫ
␤ϭ Ϫ␧ Ϫ
␥ϭ
ii i
iii
ii
1s
11s
s
⋅
(
)
(
)
⋅
(
)
⎧
⎨
⎪

⎪
⎩
⎪
⎪

(14.9)

It should be noted that if an element is to be recovered (s
i
ϭ 0), preference is
given to reusing material rather than recycling it; recycling is chosen only
when ⑀
i
is 0 (i.e., when the element is not reusable).
Again considering the index s
i
, it is appropriate to set the condition that this
can assume a value of 0 (recovery of the element) only for disassembled
elements. This equates to forcing any nondisassembled elements to be disposed
of as waste in all cases. In fact, they could be recovered only in the particular
case where they constitute a recoverable subgroup (i.e., composed of a single
recyclable material, of materials compatible with the aims of recycling, or of
elements that are all reusable, a condition that is not examined here).
The index s
i
can, instead, assume a value of 1 (disposal of the element)
independent of whether or not the corresponding element is disassembled.
It would, therefore, seem superfl uous (if not incorrect) to examine the possi-
bility of having to dispose of a disassembled element. However, this is neces-
sary since it may be advantageous to recover an element even when its

removal requires the disassembly of another element that can only be
destined for disposal, or that (although already disassembled) it is more
advantageous to consider waste.
14.5.2.3 Functions of the Costs of the Life Cycle
With regard to the cost functions, analogously to Equation (14.7), it is possible
to defi ne the end-of-life cost of element C
EoL i
:



CCc1W
cr
EoL i Prod i Dsp i i
Rcl Rcl i
iii
ii
ϭ␣ Ϫ ϩ␤ Ϫ␰
ϩϪ ␰
ᠨᠨᠨᠨ
ᠨᠨ
()
()
⎡
⎣
()
WWcW
iiDspi
i
⎤

⎦
()
ϩ␥
ᠨᠨ

(14.10)

2722_C014_r02.indd 3952722_C014_r02.indd 395 11/30/2005 1:52:50 PM11/30/2005 1:52:50 PM
© 2006 by Taylor & Francis Group, LLC
396 Product Design for the Environment
The qualitative considerations made previously for Equation (14.7) again
apply here. The only formal difference between Equations (14.10) and (14.7)
is found in the second term, which in Equation (14.10) includes the revenues
obtained from recycling the material, balancing the cost of recycling. In
Equation (14.7), there is no term of impact recovery since, as noted, it is
already included in the calculation of ei
Rcl
. For the whole system of n elements,
it is possible to express the cost functions of the entire life cycle C
LC
, or of only
the end-of-life phase C
EoL
as:



CCCCCC
LC Prod EoL Dis EoL EoL
i1

n
i1
n
ii i
ϭϩϩ ϭ
ϭϭ
()
∑∑

(14.11)

From the fi rst term of Equation (14.11), it is seen that the cost of the entire life
cycle includes the production costs for each element C
Prod i
as well as the end-
of-life costs C
EOL i
already defi ned, and the total cost of the entire process of
disassembly C
Dis
. The production costs can be expressed in a form analogous
to Equation (14.5), as a function of the quantity of material to be used and of
the more signifi cant process parameters. Alternatively, it is possible to turn to
a conventional evaluation of a component’s production costs (Ulrich and
Eppinger, 2000).
Finally, the cost of disassembly C
Dis
can be evaluated using:

C=c TSt
Dis Dis i* OR
ᠨᠨ
()

(14.12)
where c
Dis
is the disassembly cost per unit time and TS
i*
is the disassembly
time defi ned in function (14.2), calculated for the fi nal element i* of the
disassembly sequence. Since this last term expresses a time normalized with
respect to the operation of horizontal translation, it must be multiplied by the
real time t
OR
for this reference operation. In the case where it is necessary to
take into account the variable diffi culty of handling an element according to
its weight or bulk, t
OR
can be expressed as a function of the element’s weight
or volume rather than a constant value.
14.5.3 Identifi cation of the System
With regard to the identifi cation of the system by the search algorithm, also in
this second case, the formalization of the solution type to be investigated comes
down to a vector of integer numbers. Here, however, the vector is composed of
element index i, following the preliminary numeration, and the second index
corresponds to the operation typology index o
k

, relating to the operation that
must be performed to remove the i-th element. It should be noted that in this
case it is also necessary to consider a “null” operation typology.
2722_C014_r02.indd 3962722_C014_r02.indd 396 11/30/2005 1:52:51 PM11/30/2005 1:52:51 PM
© 2006 by Taylor & Francis Group, LLC
triplets of indices (Figure 14.3b). Again, the fi rst index corresponds to the
Optimal Disassembly Planning 397
The fi rst two indices are complemented by a third index, the destination
index (s) introduced in Section 14.5.2.2. Depending on whether this assumes
a value of 0 or 1, the corresponding element will be destined for recovery or
for disposal. In this case, the objective function must take into account other
signifi cant factors expressing the costs and environmental impacts of the life
cycle. This is, therefore, an example of optimization using a multiobjective
genetic algorithm (Fonseca and Fleming, 1995). Of the possible approaches to
the problem, the one chosen here is that of the weighted sum, recalling a typi-
cal formulation of conventional multiobjective optimization. The objective
function to be minimized is expressed by the weighted sum of three factors:
• The disassembly time TS, expressed by function (14.2), where the
summation is extended to all the disassembled elements
• The cost of the entire life cycle C
LC
(ignoring the term C
Dis
, the linear
function of the fi rst factor TS), or only the cost of end-of-life C
EoL

expressed by Equations (14.11)
• The environmental impact of the entire life cycle EI
LC

, or only the
impact of end-of-life EI
EoL
expressed by Equations (14.8)
This ultimately produces:



⌿ϭ␸ ϩ␸ ϩ␸
12 3
ᠨᠨ ᠨ
TS C EI
LC L
C

(14.13)

where ␸
1
, ␸
2
, and ␸
3
are the weight coeffi cients of three factors (␸
1
ϩ ␸
2
ϩ ␸
3
ϭ 1).

Given that the factors are not homogeneous, they must be normalized
before being introduced into Equation (14.13). With this aim, given a solution
to be evaluated it is suffi cient to compare the value assumed by each factor,
with the maximum value assumed by the same factor in all the solutions
under examination. A set of three normalized factors is obtained for each of
the solutions to be evaluated. The optimal solution is that corresponding to
the lowest value of the function ⌿.
It should be noted that in the context of evolutionary multiobjective opti-
mization, the weight coeffi cients also generally evolve during the process
(Gen and Cheng, 2000). However, in the formulation proposed here these
coeffi cients are excluded from the process of evolution, so that they can be
used as parameters to freely direct the investigation.
14.5.4 Generation of Disassembly Sequences and Identifi cation
of the Optimal Solution
This second tool takes advantage of the same genetic motor as the fi rst. The
generation of disassembly sequences and recovery plans is again intrinsic to
2722_C014_r02.indd 3972722_C014_r02.indd 397 11/30/2005 1:52:51 PM11/30/2005 1:52:51 PM
© 2006 by Taylor & Francis Group, LLC
398 Product Design for the Environment
the very functioning of the genetic algorithm. Each individual of each
a completely random manner, in fact represents a potential disassembly
sequence (through the fi rst two indices of each triplet) and a potential recov-
ery plan (through the third index). Also in this case, if the sequence of the
triplets respects the constraint matrices, and thus the individual represents a
truly practicable disassembly sequence, the fi tness is evaluated by the calcu-
lation of the value assumed by the multiobjective function (14.13).
When the algorithm stops, it provides the optimal (or near-optimal) solu-
tion, again expressed in the form of Figure 14.3b, consisting of a series of
element–operation–destination triplets, ordered in a way that the constraint
matrices are respected. Unlike the previous case, here the series of triplets will

not stop at just the elements to be disassembled, but will include them all. It is
not certain, however, that the optimal recovery plan will indicate the complete
disassembly of the system. As mentioned before, the tool is also designed to
defi ne the optimal level of disassembly. Elements that need not be disassem-
bled will, therefore, be identifi ed by the value assumed by the operation index
o
k
, which will assume the value corresponding to the null operation. For these
elements, the third index (s) will always have a value of 1, indicating that the
nondisassembled components can only be disposed of as waste.
14.6 Simulations and Analysis of Results
Implementing both of the proposed tools requires the use of appropriate
software. In the context of the practical experience reported here, in order to
provide a complete description of the implementation phase and of the prac-
tical use of the tools, the prototypes were developed using MATLAB
®
(The
MathWorks, Inc., Natick, MA) and tested in a series of simulations on vari-
ous mechanical systems, characterized by different architecture typologies. To
lighten the calculation, the investigation was conducted in the two-dimensional
fi eld and on assemblies consisting of a maximum number of elements limited
to 10 to 15.
On the basis of the results obtained, the fi rst prototype was found to be an
effi cient tool for identifying selective disassembly sequences, always arriving
at solutions describing feasible sequences, and often at absolute optimal
sequences. The second prototype also provides operationally feasible solu-
tions and further guides the optimal planning of disassembly and recovery,
by varying the weights given to the economic and environmental aspects in
the defi nition of the multiobjective function. Two signifi cant cases, one for
each prototype, are described in detail below. These show the analysis of the

and 5 screw fasteners).
2722_C014_r02.indd 3982722_C014_r02.indd 398 11/30/2005 1:52:51 PM11/30/2005 1:52:51 PM
© 2006 by Taylor & Francis Group, LLC
generation (represented by the vector in Figure 14.3b), although generated in
same assembly described in Figure 14.4, consisting of 11 elements (6 components
Optimal Disassembly Planning 399
Since the problem is limited to only two dimensions, the possible typologies
of disassembly operations o
k
in this case are only the four translations along
X, ϪX, Y, and ϪY, plus the null operation. As reported in the table of Figure
14.4, a value of 1.4 was associated with the vertical translations Y and ϪY
as execution times normalized with respect to horizontal translations. The
corresponding constraint matrices were compiled for each of the four possi-
ble directions of translation (the fi gure shows those relative to the two axes,
X and Y).
14.6.1 Prototype 1: Selective Disassembly
Simulations where the fi rst prototype was applied on different mechanical
assemblies generally gave excellent results. In the case under examination
(Figure 14.4), the elements were fi rst characterized through the compilation
of the vector (14.3), which associates the relevant typology with each element.
Different target elements were then selected (target of the disassembly).
5 was chosen as the target of the disassembly. These results regard fi ve different
simulations performed on the assembly, varying the total number of genera-
tions defi ned as the algorithm’s stop criterion (100, 150, 200, 250, and 350,
respectively). The solutions identifi ed as optimal are reported for each simula-
tion, with each solution consisting of an ordered series of element–operation
FIGURE 14.4 Case study: Mechanical system.
2722_C014_r02.indd 3992722_C014_r02.indd 399 11/30/2005 1:52:51 PM11/30/2005 1:52:51 PM
© 2006 by Taylor & Francis Group, LLC

Figure 14.5 shows the more signifi cant results, in the case where element