Exact methods for ALB.ppt
-
Upload
sagarbolisetti -
Category
Documents
-
view
216 -
download
0
Transcript of Exact methods for ALB.ppt
-
7/29/2019 Exact methods for ALB.ppt
1/10
Exact methods for ALB
ALB problem can be considered as a shortest
path problem
The complete graph need not be developed since
one can stop as soon as in one node all
operations have been assigned, and also because
of pruning the tree by dominance rules
The graph will still be very large -> heuristic thatconsiders only a subgraph with the cost of loosing
guarantee of optimality
-
7/29/2019 Exact methods for ALB.ppt
2/10
Pinto heuristic
1. Find some good (and feasible w.r.t. precedence) orderings
of the operations using e.g. different priority rules
2. For each of these orderings (permutations) (jl,. ... , jn) of
operations, define nodes (states)Z0 = {}, {jl}, {jl, j2}, ... , Zend = {jl, ... , jn}.
3. Draw an arrow from node ZtoZ' if Z' - Z represents a
feasible assignment of a station in the sense that cycle
time is not exceeded:4. In the resulting graph find the shortest path fromZ0 = {} to
Zend = {jl, ... , jn}.
-
7/29/2019 Exact methods for ALB.ppt
3/10
Pinto heuristic
Example: c=4
Ordering 1: 2, 1, 4, 5, 3
Ordering 2: 1, 4, 5 ,2, 3
By coincidence the optimal solution is reached!
2 1
3 4 5
1 3
2 3
2
-
7/29/2019 Exact methods for ALB.ppt
4/10
B&B algorithm by Johnson (FABLE)
DP algorithm can be considered as breadth search
All nodes in a certain stage are considered, before proceeding tothe next stage
The first complete solution is already the optimal one; if thealgorithm is stopped because of time restrictions no feasiblesolution is available.
-> B&B algorithm by Johnson treis to search the correspondingtree in the sense ofdepth search by trying to reach leaves ofthe tree soon (complete solution).
Algorithm is also known as FABLE: fast algorithm for balancing
lines effectively. Like in all B&B algorithms one tries to keep the tree small; in
addition to dominance rules also bounds are used.
-
7/29/2019 Exact methods for ALB.ppt
5/10
Branching process
In the starting node 0 no operations have been assigned
yet.
In each iteration an additional operation is assigned (or in a
backtracking step an operation is removed). A new station is opened whenever no further operation can
be assigned to the previous one (maximal stations)
In order to obtain a good first solution, the operations are
ordered according to some priority rules -> in FABLE this isdone according to the following rules (considering
precedences):
-
7/29/2019 Exact methods for ALB.ppt
6/10
Branching process
1. sort according to decreasing operation times tj (i.e.
allocate long operations first)
2. in case of a tie, use decreasing number of immediate
successors 3. in case there is still a tie choose randomly
Example:
We obtain the following ordering:1, 2, 3, 4, 5
2
1
3 4 5
1 3
2 3
2
-
7/29/2019 Exact methods for ALB.ppt
7/10
Branching process
This leads us to a leaf of the B&B tree:
Feasible solution (3 stations): {1}, {2, 4}, {3, 5}
-
7/29/2019 Exact methods for ALB.ppt
8/10
Branching process
In an exact method we must check all alternative solutions in principle.
Hence, one has to perform backtracking. We have to go back in the tree to
find the last "crossroad" where not all directions /alternatives have been
investigated yet. 1st branching opportunity: node 3 (operation 5 instead of
operation 3).
In FABLE only increasing operations number within 1 station are permitted
-> Node 6 is not to be considered any further!
-
7/29/2019 Exact methods for ALB.ppt
9/10
Branching process
Next branching opportunity: Node 1 (operation 4 instead of operation 2)
{4,2} is not allowed for station 2 (see previous slide). {4,5} is ok (no other possibilities). The
only possibility for station 3 is {2} and for station 4 {3}. Already in node 9 we see that this
cannot lead to a new optimal solution since we will not end up with < 3 stations.
Backtracking from node 9next branching opportunity: node 0.
-
7/29/2019 Exact methods for ALB.ppt
10/10
Branching process
Branching at node 0:
We try to assign operation 2 instead of operation 1. Maximal station 1 is {2} and only maximal station 2 is
{1}. In node 12 one has again 2 possible operations to be chosen, 3 and 4. We start with the lower index,
i.e. operation 3. Station {3} is maximal. In node 13 or even in node 12 one could have already stopped,since it is clear that no solution with less than 3 stations will be found. All branches have been
investigated and the algorithm stops.
The solution {1}, {2, 4}, {3, 5} with 3 stations is optimal.