You are on page 1of 8

Available online at www.sciencedirect.

com

ScienceDirect
Procedia Manufacturing 11 (2017) 949 – 956

27th International Conference on Flexible Automation and Intelligent Manufacturing, FAIM2017,


27-30 June 2017, Modena, Italy

An application of Industry 4.0 to the production of packaging films


Pierpaolo Caricatoa,*, Antonio Griecoa
a
Università del Salento, Via Monteroni, Lecce 73100, Italy

Abstract

An application of Industry 4.0 methods to the production of packaging films is presented. The production planning issues that are
addressed often include contrasting objectives and strategies between customer’s service and production optimization. We
present an Advanced Planning and Scheduling (APS) tool that allows the decision maker to automatically generate and chose
among a wide range of differently optimized scenarios. The vast amount of different results is analyzed and presented to the
decision maker, using advanced data analytics techniques in order to put him/her in the condition to rapidly take an aware and
solidly supported decision.
© 2017 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license
© 2017 The Authors. Published by Elsevier B.V.
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-reviewunder
Peer-review underresponsibility
responsibility
of of
thethe scientific
scientific committee
committee of theof27th
the 27th International
International Conference
Conference on Flexible
on Flexible Automation
Automation and and
IntelligentManufacturing
Intelligent Manufacturing.

Keywords: Automated Decision Support Systems; Manufacturing Operations; Production Planning and Scheduling

1. Introduction

The “Piano Nazionale Industria 4.0”, the Italian plan for the adoption of the Industry 4.0 paradigm by the Italian
manufacturing system, indicates a set of enabling technologies that must be used to be able to achieve the rewards
that such paradigm promises. Advanced manufacturing solutions and Big Data and analytics are among them.
We present an application of these enabling technologies to the production of packaging films, showing the
results of the application of such techniques to a real case in this sector.

* Corresponding author. Tel.: +39-0832-297806


E-mail address: pierpaolo.caricato@unisalento.it

2351-9789 © 2017 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of the scientific committee of the 27th International Conference on Flexible Automation and Intelligent Manufacturing
doi:10.1016/j.promfg.2017.07.199
950 Pierpaolo Caricato and Antonio Grieco / Procedia Manufacturing 11 (2017) 949 – 956

The production planning issues that are addressed often include contrasting objectives and strategies: on one hand
the legitimate requirement to provide the customers with an effective service, on the other hand the need to
efficiently use the production capacity. These two drivers often lead to opposite directions when a decision must be
taken.
The usage of the presented Advanced Planning and Scheduling (APS) tool allows the decision maker to rapidly
generate a wide range of different scenarios for the production planning problem at hand, that are obtained
automatically varying the weight of the different drivers defined by the user. The vast amount of different results is
then analyzed and presented to the decision maker, using advanced data analytics techniques in order to put him/her
in the condition to rapidly take an aware and solidly supported decision.
We introduce the main aspects of the production planning addressed by the presented tool, with an insight in the
artificial intelligence techniques used to represent its constraints and its objectives. We then show how different
scenarios can be built for the same problem by varying the importance given to the main defined strategies, namely:
meeting the customers’ deadlines, efficiently using the available production capacity, minimizing the stock costs.
Finally, we illustrate how the usage of an effective and reasonably compact representation of the results can rapidly
allow the user to take conscious decisions that lead to a well-balanced trade-off between the pursued contrasting
objectives.

2. The production planning problem

The production of flexible packaging films can typically include a laminating process, a coating phase, surface
treatments followed by a slitting phase. For high volume commercial film operations, indeed, production efficiency
is achieved through cast machines that produce films that are much wider than the final films required by customers:
as reported in [1], widths ranging from 2.50 to 5.00 meters are common for cast film lines, while widths of machines
for modifying films often need to be 1.80 meters or less and film widths for packaging machines are usually in the
range of 0,30 to 0,90 meters. Consequently, being able to efficiently match the several smaller widths and quantities
required by the rolls orders with the available larger widths of the reels to be slit is a key aspect in the production
planning of packaging films manufacturers.
This problem falls within the broad cutting and packing optimization area. More specifically, it can be modeled
as a variant of the CSP (Cutting Stock Problem). In the classical CSP, as introduced in 1961 by [2], a set of orders
for items of given lengths  with  has to be manufactured using larger objects of standard lengths  with
, which are supposed to be unlimitedly available. A feasibility constraint ensures that   . Each
standard length object has an associated cost and the objective is to minimize the production cost of all the required
items. A very similar problem, but with a different objective function, had been presented even earlier, in the 1957
work [3]. Here, the objective was to minimize the trim loss, i.e. the amount of spare material coming from cut
objects that is not used to manufacture items.
A common approach to solve CSP is a two-step algorithm: in step one, the set of all possible combinations of
required items, with their associated lengths, over standard length objects is determined; in step two, linear
programming is used to choose the combinations to be actually adopted for production. The most time-consuming
activity is the pattern generation phase in the first step.
We introduce a pattern generation technique that takes advantage of some problem-specific aspects that derive
from the industrial case in order to reduce the pattern generation step. We consider both technological and
commercial requirements. We also introduce an improved generation technique suitable for multiple standard
lengths problems.
The main technological aspects that can impact pattern generation are reported here as Feature 1 to Feature 4,
since what matters to the decision problem is not the specific technological aspect but how different values of each
feature can or cannot be mixed in the same production process. A commercial aspect that also impacts in the
decision is the due date associated with each order: orders associated with deadlines too distant in time cannot be
combined.
We consider a set of orders for reels of packaging films to be manufactured starting from wider rolls of input
material. Each order is characterized by:
Pierpaolo Caricato and Antonio Grieco / Procedia Manufacturing 11 (2017) 949 – 956 951

• the specific material to be used;


• the number of reels to be produced, expressed in terms of an ordered quantity, a positive tolerance and a negative
tolerance on such quantity;
• the width of the reels to be produced (TBP);
• the length of each reel;
• the required internal diameter;
• four technical features required by the order;
• the due date of the order.
Raw material rolls are available in standard widths and the number of reels available for each width is virtually
unlimited for the problem at hand. All orders must be manufactured, respecting the specific requirements for each
order, while a cost function considering both the generated spare material and the number of used rolls has to be
minimized. The considered problem can be framed within the Dyckhoff typology introduced in [4] through a
decomposition approach.

2.1. Problem decomposition

The problem can be decomposed into smaller independent sub-problems and each sub-problem can be iteratively
addressed as a cutting stock problem with technological constraints.
Some slitting machines only allow the combined production of orders that share the exact same technological
features and the same required length, while some other machines allow for a combination of at most two different
values of some features and length, corresponding with the ability to wind the produced reels using the two sides of
the machine.
Hence, for the former machines, each possible different combination of the four technological features plus the
required length lead to an independent sub-problem, while, on the latter machines, only the first feature is exclusive,
while the other three features and the required length can be combined in order to optimize the production, leading
to less but larger sub-problems, one for each value of the first technological feature.
The features that lead to the definition of the sub-problems can be considered as fixed, in the sense that, once the
sub-problems have been defined, they can be completely ignored by the dynamic pattern generation procedure,
unlike the remaining variable ones, that are directly addressed and used by such procedure. In the remainder of the
work we will only refer to the latter, more complex machine types.

2.2. Problem classification

Each individual so defined problem can be considered as a variant of the classical CSP, identified as 1/V/D/R in
the Dyckhoff typology. The single dimension (“1”) refers to the width of the items to be cut, while the “V” means
that all orders must be produced by making a selection among the different width rolls available (hence the “D” in
the typology). In the test case, there exist relatively few possible widths, while the availability for each roll width is
virtually infinite. According to the typology, this is defined by the letter “R”.
Actually, for the machines that allow the combination of up to two different reels required lengths in a same roll,
the dimension is somewhere in the middle between a pure 1D and a pure 2D problem, since the different lengths
lead to a two-dimensional pattern. However, the very limited possible combinations of different lengths lead to a
treatment of this aspect in the pattern generation, as detailed in the following sections, that does not differ too much
from a mono-dimensional approach.

3. The proposed tool

The firm information system asynchronously publishes dynamic data regarding the orders to be planned,
updating them anytime a new order is acquired/confirmed or a partial production progress is available for an already
planned order.
952 Pierpaolo Caricato and Antonio Grieco / Procedia Manufacturing 11 (2017) 949 – 956

The proposed tool accesses this information when the user decides to make a new production plan, including new
orders and yet to be planned quantities of previous orders. The tool’s UI (User Interface) allows the user to quickly
review the available dynamic data (see Fig. 1), but it also allows the user to specify configuration data, used to
specify the machines’ characteristics and decision-related parameters such as the weights of different decision
drivers in the proposed scenarios (see Fig. 2).

Fig. 1. Dynamic orders data.

Fig. 2. Configuration data.


Pierpaolo Caricato and Antonio Grieco / Procedia Manufacturing 11 (2017) 949 – 956 953

Once the user has defined the parameters to be considered in the current scenario, the tool applies the problem
decomposition approach described in section 2.1 to define independent decision problems that are hence addressed
using a two-steps approach: in the former phase, a set of feasible patterns is generated, while the latter step uses an
ILP (Integer Linear Programming) model to find the best solution according to the user-defined parameters.

3.1. Feasible patterns

A pattern is a combination of reels to be cut from a input roll of raw material. The number of times each pattern
is applied is called the pattern frequency. A pattern is feasible for the confronted problem only if it matches all the
features required by the orders to be cut from it.
One of the major issues to be addressed in a two steps approach is the large number of feasible patterns that can
be generated with relatively small problems. Pattern generation for cutting stock problems has been studied in many
works in the literature. Patterns are commonly generated in two ways: in advance or on-line. With advance column,
all of the feasible patterns are generated: it can be used for full pattern generation in small/medium problems or
when a representative subset is generated for a large problem. On-line pattern generation is more suitable for large
integer problems: it uses a column generation technique analogous to the used in the classic one-dimensional CSP
[5]. A heuristic pattern generation procedure called SHP (Sequential Heuristic Procedure) was first introduced in [6]
and [7]; the procedure was improved in later works such as [8] and [9].
We adopted a simple pattern generating procedure based on an ad hoc solution method described in literature for
the knapsack problem, as described in [10]. We adapted this approach in order to take into consideration, and also to
take advantage of, the following problem-specific aspects: for each variable feature, only one or two different values
can be present within a same pattern; different required lengths can be included in the same pattern only if they are
exact multiples of each other. An example described in the results section 3.1 better clarifies this aspect.
Another feasibility check that allows to generate less patterns is the maximum gap allowed by the user between
the earliest and latest due dates among all the orders included in a pattern.

3.2. ILP model

Once the feasible patterns have been generated for each sub-problem, an Integer Linear Programming (ILP)
model is used to select which patterns are to be used, and in which quantity, for the production of the required reels.
The objective is the minimization of the total amount of trim on the input rolls used in the production plan, as well
as the number of different patterns used, according to the weights the user provides for these two objectives.
Let us denote with  a generic pattern and with  a generic coil type. The following quantities are used in the ILP
model:
•  is the set of all the generated feasible patterns;
•      is the set of the reels orders, partitioned into mandatory and optional orders: an order is optional if its
due date falls beyond the limit date specified by the user (see Fig. 3);
•    is the integer decision variable related to pattern   , indicating how many replicas of such pattern must
be produced, if any;
•  is the boolean decision variable related to pattern   , indicating if such pattern must be produced or not;
•  is the quantity of reels of type  contained in pattern   , including both the multiplicity of  along the width
and along the length of the roll;
•  and  are the minimum/maximum quantity of reels of type  that must be produced;
•  is the trim for pattern , i.e. the unused part of the roll width in the pattern;
•  is the weight specified by the user for the trim minimization objective, while  is the weight for the number of
different patterns used minimization objective.
Given the above notation, the mathematical model used for the production planning problem is the following:
954 Pierpaolo Caricato and Antonio Grieco / Procedia Manufacturing 11 (2017) 949 – 956

       (1)
 
subject to
       (2)


       (3)



    (4)
 
 
      (5)
      (6)
      (7)

The objective function (1) guarantees that the optimal value in chosen, according to the weights defined by the
user. Constraints (2) guarantee that the maximum quantity required for each order is not exceeded, while constraints
(3) guarantee that at least the minimum quantity required by each order is produced, but only for the mandatory
orders. Constraints (4) and (5) guarantee logical coherence between x and y variables. Finally, constraints (6) and (7)
define the domains of the decision variables.

3.3. Results

The ILP model is solved using standard MIP (Mixed Integer Programming) solvers: we tested it both with
leading commercial solvers (IBM CPLEX and Gurobi) and with an open source solver (Coin-OR CBC), and all the
real instances of the sub-problems were solved to optimality within a 300 seconds time limit.
The screenshot in Fig. 3 shows an example of the results proposed by the tool to the user. The rows in the upper
data grid represent different sub-problems, as derived from the decomposition detailed in section 2.1, while the rows
in the data grid below show the proposed patterns to be produced for the selected sub-problem. The content of the
column “Machine” for the highlighted row shows an example of a combination, within a same pattern, of orders
with different lengths:
• the entry “2x1080x1x13500@85” means that the pattern includes two reels of the order identified by the number
85 (which requires a 1080 cm width and a 13500 meters length) combined along the width, for all the length of
the input roll;
• the entry “1x815x2x6750@67”, on the other hand, means that the pattern contains two reels of the order
identified by the number 67, which requires a 67 cm width, combined along the length;
• the total width required is 2975 cm (column “Used width”), while the smallest available input roll provides a
3040 cm width (column “Input width”), of which only 2980 actually usable after the technically unavoidable trim
(column “Available width”), leading to a calculated trim of 5 cm (column “Net trim”), corresponding to a 0.17%
percentage trim (column “Net % trim),
• the frequency associated with this pattern, i.e. the number of times it has to be produced, is 3 (column
“Quantity”);
• finally, the “First date” and “Last date” columns report the earliest and latest due dates of the orders included in
the pattern.
The pattern is feasible because the length of the first order is exactly the double of the length of the second order.
Naturally, this optimization leads to good results in the addressed industrial case because order lengths are only
possible for standardized lengths.
The overall performance indicators for the selected sub-problem are reported at the bottom right of the window,
showing the total weight of the input rolls (field “Produced KGs”), the total weight of this production that is gross
trim, including technical trim (field “Trim KGs”) and the percentage of gross trim (field “Trim %”).
Pierpaolo Caricato and Antonio Grieco / Procedia Manufacturing 11 (2017) 949 – 956 955

Fig. 3. Results.

4. Usage scenarios

The proposed tool has been introduced to support production planning related decision-makers, as an application
of the Industry 4.0 renewal process that is taking place. In particular, the ability to be constantly updated with the
dynamic information coming both from the sales department, when new orders are confirmed to the customers, and
from the flow shop, when parts of orders in production are completed, is an achievement made possible through the
communication infrastructure developed in the firm.
The tool takes advantage of the availability of such data, but also introduces sophisticated data analytics and
artificial intelligence techniques that allow the user to rapidly check the results of different strategic or tactical
decisions on the quality of the achievable results.
For instance, a key aspect, related with the availability of more or less storage space in the particular moment, is
the ability for the user to define how far to go into the analysis of orders with due dates in the future, by setting
different values for the “limit date” parameter (as seen in Fig. 3 and detailed in the 3.2 section). The fast processing
of the data and the production of the best solution achievable according to the choice made, allow the user to have
an immediate feedback of the effects of his/her choice in terms of production efficiency.
Different values of the “Max days gap in a pattern”, as well, which is used in the definition of the feasible
patterns to consider, allow the user to experiment the effects of the possibility to mix more or less the combined
production on the production efficiency, letting him/her have actual, proved performance indicators to decide when
this gap can be effectively increased.

5. Conclusion

An application of Industry 4.0 techniques to the production planning workflow in a firm that produces packaging
films. The large availability of always up-to-date data and the ability to effectively process them with advanced
planning techniques proved to be worthy to help the user take better and more numerically provable decisions. A
further development of this work is the design and implementation of a more advanced and interactive dashboard to
allow the user to collect and better interpret results from the different tested scenarios.
956 Pierpaolo Caricato and Antonio Grieco / Procedia Manufacturing 11 (2017) 949 – 956

Acknowledgements

This work was supported by Progetto PON/MIUR Vis4Factory: Sistemi Informativi Visuali per i processi di
fabbrica nel settore dei trasporti - Codice Progetto: PON02_00634_3551288

References

[1] W. A. Jenkins and K. R. Osborn. Plastic Films: Technology and Packaging Applications. CRC Press, 1992.
[2] P. C. Gilmore and R. E. Gomory: “A linear programming approach to the cutting-stock problem”, Operations Research, Vol. 9, No. 6, pp.
849-859, 1961.
[3] K. Eisemann: “The trim problem”, Management Science, Vol. 3, No. 3, pp. 279-284,1957.
[4] H. Dyckhoff: “A typology of cutting and packing problems”, European Journal of Operational Research, Vol. 44, No. 2, pp.145-159, 1990.
[5] M P. Johnson, C. Rennick and E. Zak, “Skiving addition to the cutting stock problem in the paper industry”, Journal of Siam Review, Vol. 39,
No. 3, pp. 472-483.
R. W. Haessler: “A heuristic programming solution to a nonlinear cutting stock problem”, Management Science, Vol. 17, No. 12, pp. 793-802,
1971.
[7] R. W. Haessler: “Controlling cutting pattern changes in one-dimensional trim problems”, Operations Research, Vol. 23, No. 3, pp. 483–493,
1975.
[8] R. Vahrenkamp, “Random search in the one-dimensional cutting stock problem”, European Journal of Operational Research, Vol. 95, pp.
191–200, 1996.
[9] S. Umetani, M. Yagiura and T. Ibaraki: “One-dimensional cutting stock problem to minimize the number of different patterns”, European
Journal of Operational Research, Vol. 146, No. 2, pp. 388-402, 2003.
[10] S. M. A. Suliman. “Pattern generating procedure for the cutting stock problem”, International Journal of Produc-tion Economics, Vol. 74,
No. 3, pp. 293-301, 2001.

You might also like