Bin Packing and Cutting Stock Problems: Mathematical Models and Exact Algorithms

Bin Packing and Cutting Stock Problems: Mathematical
Models and Exact Algorithms
Maxence Delorme(1) , Manuel Iori(2) , Silvano Martello(1)
(1) DEI, University of Bologna, Italy,

(2) DISMI, University of Modena and Reggio Emilia, Italy
BOLOGNA 2015
Delorme, Iori, Martello Bin Packing and Cutting Stock Problems BOLOGNA 2015 1 / 35
Outline
1 Introduction
2 Most common exact methods for solving the BPP and the CSP
3 Computational Results
4 Conclusion
Introduction
The Bin Packing Problem (BPP)
Classical Bin Packing Problem

Given a set of weighted items and an unlimited number of identical capac-
itated bins, the Bin Packing Problem consists in packing all the items into
the minimum number of bins.
Bins
Introduction

Items Bins
9 9 9
4 4
3 3
2 2
Introduction

Items Bins
2 2
3 3
9
4 4 4 4
3 3
2 2
Introduction
The Cutting Stock Problem (CSP)
Classical Cutting Stock Problem

Given a set of order requirements, each requirement consisting in a demand
and a width, and an unlimited number of identical rolls, the classical cutting
stock problem consists of determining the smallest number of rolls that have
to be cut in order to satisfy all the demands.
Bins
Introduction

Order requirements Rolls
2 x2 9
3 x2 9
4 x2 9
Introduction

Order requirements Rolls
2 x2 4 3 2
3 x2 4 3 2
4 x2 9
Introduction
Some applications for the BPP and the CSP
I Cutting materials in industry (wood, paper, aluminium ...)

I Kantorovich, Mathematical methods of organizing and planning
production, Management Science, 1960 (originally from 1939)
I Stadler, A one-dimensional cutting stock problem in the aluminium
industry and its solution, European Journal of Operational Research,
1990
I Loading when 1 dimension is considered (file storage on computers,
container loading ...)
I Solving more difficult problems (vehicle routing, multi-dimensional
BPP, ...)
Introduction
Context of our work

I BPP and CSP are more and more studied by researchers
The subject is of interest
I Many techniques can be used to solve the BPP and the CSP
A survey is relevant
I All the literature instances are solved
A will to create a new computational challenge
Introduction
Objectives of our work
I Gather in a survey the most important articles

I Test the efficiency of some exact methods proposed in the literature
(Branch-and-Bound, Branch-and-Price, Pseudo-Polynomial models
...)
I Study the behaviour of those methods when the parameters of the
test instances change
I Propose new instances that are difficult to solve in practice
Most common exact methods for solving the BPP and the CSP
The textbook BPP model solved by ILP solver
Martello, Toth, Knapsack Problems: Algorithms and Computer Implementations, Wiley, 1990
Roots in the seminal work by Kantorovich, Mathematical methods of organizing and planning
production, Management Science, 1960 (originally from 1939)
m
X
min yi (1)
i=1
Xn
s.t. wj xij ≤ cyi (i = 1, . . . , m), (2)
j=1
m
X
xij = 1 (j = 1, . . . , n), (3)
i=1
yi ∈ {0, 1} (i = 1, . . . , m), (4)

xij ∈ {0, 1} (i = 1, . . . , m; j = 1, . . . , n). (5)
Easy to implement, may be efficient on small instances (n ≤ 100)
Branch & Bound algorithms
Martello, Toth, Knapsack Problems: Algorithms and Computer Implementations, Wiley, 1990
Scholl, Klein, Jürgens, BISON: a fast hybrid procedure for exactly solving the one-dimensional
BPP, Computers & Operations Research, 1997
Main idea: Explore (in a ”smart” way) through an enumeration tree all the
feasible packings
Smart because:
I Use of reduction procedures
I Use of a set of specially designed Lower Bounds (L1, L2, L3 ... L6)
I Use of fast and good heuristics to get Upper Bounds
Avoid the use of solvers. Generally good on small instances (n ≤ 100), or

when the preprocessing is efficient.
Branch & Price algorithms

Gilmore, Gomory, A linear programming approach to the CSP, Operations research, 1963
Vance, Barnhart, Johnson, Nemhauser, Solving binary CSP by column generation and
branch-and-bound, Computational optimization and applications, 1994
Belov, Scheithauer, A branch-and-cut-and-price algorithm for one-dimensional stock cutting and
two-dimensional two-stage cutting, European Journal of Operational Research, 2006
Main idea: Consider the Bin Packing Problem as a Set Covering Problem
Set Covering Problem (SCP)

Given a set of elements {1, 2, ..., n} (called the universe) and a family S of m sets whose union
equals the universe, the set covering problem is to identify the smallest subfamily of S whose
union equals the universe.
I The elements {1, 2, ..., n} are the items
I The family S is composed by every feasible bin (column generation used to generate bins)
Very good when the lower bound is equal to the optimum (true in 99% of the case), conjectured
to be at most one bin away from the optimum (called Non-IRUP).
Constraints Programming
Shaw, A constraint for bin packing, 2004

Schaus, Régin, Van Schaeren, Dullaert, Raa, Cardinality reasoning for bin-packing constraint:
Application to a tank allocation problem, 2012
Main idea: Use properties of feasible solutions to define the search domain.
Shaw did a very good job and created a constraint implemented in CPLEX.
Can be efficient on small instances (n ≤ 100) and has the advantage of

easily allowing additional constraints.
Pseudo Polynomial Models solved by ILP solver

Dyckhoff. A new linear programming approach to the cutting stock problem, Operations
Research 1981
Valerio de Carvalho, Exact solution of bin-packing problems using column generation and
branch-and-bound, Annals of Operations Research 1999
Cambazard, O’Sullivan, Propagating the bin packing constraint using linear programming, 2010
Brandão, Pedroso, Bin Packing and Related Problems: General Arc-flow Formulation with
Graph Compression, 2013
Main idea: Consider a bin as a path in a graph where arcs are items
Arcs

Research 1981
Items Full graph
2 x2
3 x2 0 1 2 3 4 5 6 7 8 9
4 x2

Research 1981
Items Full graph
2 x2
3 x2 0 1 2 3 4 5 6 7 8 9
4 x2

Research 1981
Items Full graph
2 x2
3 x2 0 1 2 3 4 5 6 7 8 9
4 x2

Research 1981
Items Full graph
2 x2
3 x2 0 1 2 3 4 5 6 7 8 9
4 x2

Research 1981
Items Full graph: 30 arcs, 10 nodes
2 x2
3 x2 0 1 2 3 4 5 6 7 8 9
4 x2

Research 1981
Items Optimal solution
2 x2
1 1 2
3 x2 0 1 2 3 4 5 6 7 8 9
1 1
4 x2

Research 1981
Items Reduced graph
2 x2
3 x2 0
4 x2

Research 1981
Items Reduced graph
2 x2
3 x2 0 4 8
4 x2

Research 1981
Items Reduced graph
2 x2
3 x2 0 3 4 6 7 8
4 x2

Research 1981
Items Reduced graph
2 x2
3 x2 0 2 3 4 5 6 7 8 9
4 x2

Research 1981
Items Reduced graph: 19 arcs, 9 nodes
2 x2
3 x2 0 2 3 4 5 6 7 8 9
4 x2

Research 1981
Items Optimal solution
2 x2
2 2
3 x2 0 2 3 4 5 6 7 8 9
2
4 x2
Model
min z (6)

X X  z if e = 0;
s.t. − xde + xef = −z for e = c; (7)
0 otherwise,

d∈δ − (e) f ∈δ + (e)
X
xd,d+wi ≥ bi (i = 1, ..., m), (8)
(d,d+wi )∈A0
xde ≥ 0 and integer (d, e) ∈ A0 , (9)

− +
where δ (e) (resp. δ (e)) denotes the set of arcs entering (resp. emanating from) e.
Very good when the lower bound is equal to the optimum and/or when the number of
arcs is not too big (c < 1000) .
Computational Results
Instances
I Literature instances (1615)
I 1210 instances from Sholl, Klein and Juergens (1997)
I 28 instances from Schoenfield (2002)
I 200 instances from Schwerin and Waescher (1998).
I 17 instances from Waescher and Gau (1996).
I 160 instances from Falkenauer (1996).
I Randomly generated instances (3840)

I Variable number of items: 50,...,1000
I Variable bin capacity: 50,...,1000
I Variable parameters for the distribution of the items
I 10 instances for each combination
I Augmented Non-IRUP (ANI) instances (250 + 250 Augmented IRUP (AI))

I Variable number of items: 201, 402, 600, 801, and 1002
I Each have its own maximum capacity: 2 500, 10 000, 20 000, 40 000, and 80 000
I 50 instances for each of the 5 sets
I An AI instance was created from each ANI instances by splitting 1 item so that the Non-IRUP is lost
Results for literature instances
Number of literature instances (average gap wrt lower bound) solved in

less than one minute, on an Intel Xeon 3.10GHz with 8GB of RAM
I Best methods seem to be the B&P algorithm of Belov and the

Pseudo Polynomial models
I All instances can be solved in less than 10 minutes by BELOV
Results for randomly generated instances
Number of random instances solved in less than one minute (average gap
wrt lower bound) when varying n, on an Intel Xeon 3.10GHz with 8GB of
RAM
I No change in the efficiency of the methods

I No difficulty for solving random instances even with “large” capacity
and number of items
Results for ANI and AI instances
Number of difficult instances (ANI) solved in less than 1 hour (average

gap wrt lower bound), on an Intel Xeon 3.10GHz with 8GB of RAM
I No change in the efficiency of the methods

I ANI instance are indeed hard to solve
Results for a subset of instances
Number of selected instances solved [average time in seconds] using

different version of CPLEX, on an Intel Xeon 2.66GHz with 24 GB of RAM
I Efficiency of pseudo-polynomial models highly depend on the

performance of the ILP solver used
Conclusion
Conclusion
Conclusion
I According to our results, the best methods to solve the BPP and the
CSP seem to be ARCFLOW, VPSOLVER and the B&P algorithm of
Belov et al.
I The use of pseudo-polynomial models for solving the BPP became
relevant thanks to the performance of the ILP solver
I There are some properties that make an instance difficult for the tool
we tested
I Some new BPP instances are “open” !

Bin Packing and Cutting Stock Problems: Mathematical Models and Exact Algorithms

Uploaded by

Copyright:

Available Formats

Bin Packing and Cutting Stock Problems: Mathematical Models and Exact Algorithms

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Bin Packing and Cutting Stock Problems: Mathematical Models and Exact Algorithms

Uploaded by

Copyright:

Available Formats

Bin Packing and Cutting Stock Problems: Mathematical

Models and Exact Algorithms

Maxence Delorme(1) , Manuel Iori(2) , Silvano Martello(1)

(1) DEI, University of Bologna, Italy,

The Bin Packing Problem (BPP)

Classical Bin Packing Problem

The Bin Packing Problem (BPP)

Classical Bin Packing Problem

The Bin Packing Problem (BPP)

Classical Bin Packing Problem

The Cutting Stock Problem (CSP)

Classical Cutting Stock Problem

The Cutting Stock Problem (CSP)

Classical Cutting Stock Problem

Order requirements Rolls

The Cutting Stock Problem (CSP)

Classical Cutting Stock Problem

Order requirements Rolls

Some applications for the BPP and the CSP

I Cutting materials in industry (wood, paper, aluminium ...)

Context of our work

Objectives of our work

I Gather in a survey the most important articles

The textbook BPP model solved by ILP solver

yi ∈ {0, 1} (i = 1, . . . , m), (4)

Easy to implement, may be efficient on small instances (n ≤ 100)

Branch & Bound algorithms

Avoid the use of solvers. Generally good on small instances (n ≤ 100), or

Branch & Price algorithms

Set Covering Problem (SCP)

I The elements {1, 2, ..., n} are the items

Shaw, A constraint for bin packing, 2004

Can be efficient on small instances (n ≤ 100) and has the advantage of

Pseudo Polynomial Models solved by ILP solver

Pseudo Polynomial Models solved by ILP solver

Items Full graph

Pseudo Polynomial Models solved by ILP solver

Items Full graph

Pseudo Polynomial Models solved by ILP solver

Items Full graph

Pseudo Polynomial Models solved by ILP solver

Items Full graph

Pseudo Polynomial Models solved by ILP solver

Items Full graph: 30 arcs, 10 nodes

Pseudo Polynomial Models solved by ILP solver

Items Optimal solution

Pseudo Polynomial Models solved by ILP solver

Items Reduced graph

Pseudo Polynomial Models solved by ILP solver

Items Reduced graph

Pseudo Polynomial Models solved by ILP solver

Items Reduced graph

Pseudo Polynomial Models solved by ILP solver

Items Reduced graph

Pseudo Polynomial Models solved by ILP solver

Items Reduced graph: 19 arcs, 9 nodes

Pseudo Polynomial Models solved by ILP solver

Items Optimal solution

Pseudo Polynomial Models solved by ILP solver