You are on page 1of 23

A simulated annealing

algorithm for solving the bi-


objective facility layout
problem
Introduction
• The optimal design of physical departments is one of the most important
issues to be considered in the design of a manufacturing system.

• The material handling cost which comprises between 20% and 50% of the
total operating expenses within manufacturing is the most significant
measure for determining the efficiency of a layout.

Traditionally, there are two approaches for FLP.

• Quantitative approach aiming at minimizing the total material handling cost


between the departments
• Qualitative approach aiming at maximizing closeness rating scores between
departments based on a closeness function.
Disadvantages

•Qualitative approach is based on pre-assigned numerical values for different closeness


ratings but does not consider the materials handling cost due to work flow between the
departments.

• The quantitative approach does not consider the qualitative objectives. The layout
designer may have to keep certain departments closer while other departments are to be
kept further apart.

Solution Approach

•In order to do a more effective layout, both objectives should be considered can be
done by adopting a bi-objective approach for facility layout problem.

•SA algorithm is presented minimizing the objective function that combine the total
materials handling cost and the closeness rating score.
Literature review
Facility layout problem is formulated as quadratic assignment problem (QAP).
Consider the problem of allocating N facilities to N locations, with the cost
being a function of the distance and flow between the facilities.

QAP is one of the hardest optimization problems and no exact algorithm can
solve problems of size N > 30 in reasonable computational time.

For multi objective facility layout problem (MOFLP), the quadratic assignment
formulation is shown in Eqs. (1)–(4) in Sha and Chen (2001)
Where,
Xij = 1 if facility i is assigned to location j;
• 0 otherwise.
Aijkl the cost of locating facility i at location j and facility k at location l.

Eqn (1) It represents the combination of the total materials handling cost and
the closeness rating score in MOFLP formulation.

Eqs. (2) and (3) ensure that each location is assigned only one facility and each
facility is assigned to only one physical location, respectively.

Numerous methods have been suggested in the literature up to now in order to


solve the MOFLP.

These studies can be assigned into four different groups according to their ways
of combining the objectives;
Rosenblatt defines the parameter Aijkl as Aijkl = w2Cijkl -w1Rijkl. Here, Cijkl represents
(1979)
the total material handling cost, and Rijkl represents the total closeness rating
score. w1 and w2 are the weight used for the objectives.

Fortenberry sets Aijkl = fikdjlrik. It is named as multiplicity model. In this expression f ik shows
and Cox
(1985)
the handling of material between i and k facilities, r ik shows the closeness
rating score between i and k facilities, and djl shows the distance
between locations j and l.
Urban Aijkl = djl(fik + crik) is used where parameters have the same meaning as those
(1987)
used in Fortenberry and Cox (1985). The additional parameter c is a constant
which states the importance of closeness rating score due to handling of
material. C is taken as equal to the highest level of material handling
between facilities.
Khare, Khare, Defined Aijkl term as Aijkl = w1djlfik + w2rikdjl. The definition of the parameters
and Neema
(1988)
are same as given before, and w1 and w2 are the weights
used to unify the two objectives, respectively.
Deficiencies

All factors may not be represented on the same scale: for example, values for work
flow may range from zero to a tremendous amount, while closeness rating values
may range from 1 to 4. As a result, the closeness ratings would be dominated by
work flow and have little impact on the final layout.
•Measurement units used for the objectives may be incomparable:
The total closeness rating score is only an ordinal value; on the other hand, the
material flow handling is measured according to cost. Aggregation of these two
values with different measurement units in an algebraic operation is unsuitable.

Harmonosky and Tothero (1992) suggests an approach that normalizes all


objectives, before combining them.
To normalize an objective, each relationship value is divided by the sum of all
relationship values for that objective. Eq. (5) is used to do this

where Sikm represents the relationship value between i and k for objective m and
Tikm represents the normalized relationship value between i and k for objective m.
Next, all values are multiplied by the weights representing the relative
importance of each objective m (wm). Then, the total of all values for each pair
of departments is calculated.

Where wm is the weight for objective m, T is the number of objectives, N is the


number of departments, dij is the distance between locations j and l.
Xij = 1 if department i is assigned to location j or, 0 otherwise.
i and k are indices for the department numbers and j and l are indices for the
location numbers.
Simulated annealing for the MOFLP
• SA is a stochastic search method, which imitates the physical annealing of solid,
for finding solution to combinatorial optimization problems.

• SA has an advantage over the other meta-heuristic algorithms in terms of the ease
of implementation and gives reasonably good solutions for many combinatorial
problems.

step procedure
Step 1 Read the input data (normalized flow matrix, size of the problem,
normalized relationship matrix, the weights for two objectives) and the
parameters of simulated annealing (Tin = initial temperature, a = cooling
rate, NIET = the number of trials to be performed with the
same temperature value, iteration number).
Step 2 Start temperature counter: el = 0
Step 3 Create a random initial solution (S0) and calculate the weighted cost of
initial solution (E0) using the Eq. (6).
Sbest = Sc = S0; Ebest = Ec = E0
Step 4 Make the iteration counter 0 at each temperature level: il = 0
Annealing schedule
• In general, SA algorithms start with a randomly generated initial solution or
with a solution produced using a heuristic. In this work, we use a randomly
generated initial solution.
• In the proposed SA, the search moves from the current solution to a
neighbouring solution by swapping two departments. Two departments are
selected randomly and these departments are swapped.
• The probability of acceptance is defined as the probability of accepting a non-
improving solution. This is determined based on the following probability:
P(E)= exp(-E/T);

where T is the temperature and DE represents the change in the cost of the
neighbouring solution and the cost of the current solution.

• If x is a randomly generated number between 0 and 1, and x < P(E), then


accept the non-improving neighbouring solution as the current solution.
Otherwise, reject the non-improving neighbouring solution, and keep the
current solution.
• The value of the temperature at the beginning of the schedule should be large
enough so that most of the initial movements are accepted. However, as the
temperature is reduced, the probability of accepting a non improving
solution reduces.

• For a good solution the decrease in temperature must be as fast as possible


to be computationally efficient, but slow enough to maintain equilibrium
within the system.
The most common cooling function uses a geometric decrement function
(Tk = Tk-1). Here a is cooling rate and its value is between 0 and 1 but it is
close to 1.

• To finalize the algorithm various methods are used. For example some stops
the iteration s when the total iteration number reaches, a specified value
while other stop when there is no accepted move in a given number of trials.

• In the concerned experiment the search is ended when we reach the


specified maximum number of iterations (el max).
Result
Algorithm is tested on two sets of test problems.
• Set 1: contains two problems given by Harmonosky and Tot- hero (1992)
which consist of 8 and 12 departments
• Set 2: contains four problems given by Chen and Sha (1999) which consist
of 8, 12, 15, and 20 departments.

Two comparisons are made for the mentioned two sets.


1. The first data set results obtained by the proposed SA algorithm are
compared with H & T (1992) and C & S (1999) results.
2. The second data set results are compared with C & S (1999) results.
SET 1
• For 8 weight combination: the proposed SA algorithm and Chen and Sha’s
procedure produced same results while
• For 12 weight combinations: the proposed SA algorithm clearly
outperformed the other two procedures.
SET 2
• For 8 weight combination: the proposed SA algorithm provide better results
than C & S procedure for six weight combinations. Reduction in the cost are
quiet significant for weights 0.5, 0.6, and 0.7.
• For 12 weight combinations: the proposed SA algorithm procedure produced
same results for weights 0.0, 0.1, and 0.9 and for the rest, the proposed SA
algorithm is slightly better than C & S procedure .
• For 15 weight combination: for the seven weight combination the proposed
SA algorithm obtained slightly better results than C & S procedure and for
the rest the results are same.
• For 18 weight combinations: the proposed SA algorithm obtained better
results for all weight combinations
Conclusion
• The proposed SA algorithm for BOFLP has obtained the best results for the
most weight combination for each problem.
• For the rest of the weight combination, the proposed SA algorithm has
obtained the same results with Harmonosky and Tothero’s (1992) results and
Chen and Sha’s (1999) results.
• Above results proved that the proposed SA algorithm performed better than
two other heuristics for all test problems according to solution quality.
Future Work
• For future studies, using other meta-heuristic methods (genetic algorithm,
tabu search, ant colony algorithm, etc.) or employing hybrid methods can be
thought.
• Also, it will be suitable to search solutions to problems where the data is
dynamic, fuzzy or stochastic
References
• Armour, G. C., & Buffa, E. S. (1963). A heuristic algorithm and simulation
approach to relative allocation of facilities. Management Science, 9, 294–309.
• Buffa, E. S., Armour, G. C., & Vollmann, T. E. (1964). Allocating facilities
with CRAFT. Harvard Business Review, 42, 136–158.
• Chen, C.-W., & Sha, D. Y. (1999). A design approach to the multi-objective
facility layout problem. International Journal of Production Research, 37(5),
1175–1196.
• Chen, C.-W., & Sha, D. Y. (2005). Heuristic approach for solving the multi-
objective facility layout problem. International Journal of Production
Research, 43(21), 4493–4507.
• Czyzak, P., & Jaszkiewicz, A. (1998). Pareto simulated annealing – A
metaheuristic technique for multiple-objective combinatorial optimization.
Journal of Multi-Criteria Decision Analysis, 7, 34–47.
• Dutta, K. N., & Sahu, S. (1982). A multigoal heuristic for facilities design
problems: MUGHAL. International Journal of Production Research, 20(2),
147–154.
• Khare, V. K., Khare, M. K., & Neema, M. L. (1988). Combined computer-
aided approach for the facilities design problem and estimation of the
distribution parameter in the case of multigoal optimization. Computers and
Industrial Engineering, 14(4), 465–476.
• Kirkpatrick, S., Gelatt, C. D., Jr., & Vecchi, M. P. (1983). Optimisation by
simulated annealing. Science, 220(4598), 671–680.
• Malakooti, B., & Tsurushima, A. (1989). An expert system using priorities
for solving multiple-criteria facility layout problems. International Journal of
Production Research, 27(5), 793–808.
• Muther, R. (1974). Systematic layout planning (2nd ed.). Boston: Cahners
Books.
• Rosenblatt, M. J. (1979). The facilities layout problem: A multi-goal
approach. International Journal of Production Research, 17(4), 323–332.
• Sahni, S., & Gonzales, T. (1976). P-complete approximation problems.
Journal of ACM, 23(3), 555–565
• Serafini, P. (1994). Simulated annealing for multiple objective optimization
problems. In G. H. Tzeng et al. (Eds.). Multiple criteria decision making:
Expand and enrich the domains of thinking and application (Vol. 283).
Springer-Verlag.
THANK YOU

You might also like