Professional Documents
Culture Documents
ScienceDirect
Available online at www.sciencedirect.com
ScienceDirect
Procedia Manufacturing 00 (2019) 000–000
www.elsevier.com/locate/procedia
10th CIRP Sponsored Conference on Digital Enterprise Technologies (DET 2020) – Digital Technologies as
Enablers of Industrial Competitiveness and Sustainability
Abstract
Since the appearance of computers, engineers recognized their potential. Due to this fact, engineers are utilizing computer usage during their
daily tasks. When it comes to analysis, simulation is one of the most valuable tools an engineer can apply. In this paper a mixed integer
programming-based description and a discrete event-based model is presented. The mathematical approach is used for rough checking, while the
more detailed event-based simulation is used to examine the effects of certain parameters of the assembly cell. Based on the results, the bottleneck
of the line and the possibly over-capacitated buffers were identified. Furthermore, a worker regrouping strategy is determined with the help of
the models.
© 2021The
© 2020 TheAuthors.
Authors. Published
Published by Elsevier
by Elsevier B.V.B.V.
This
This isisan
anopen
openaccess
access article
article under
under the BY-NC-ND
the CC CC BY-NC-ND licenselicense (http://creativecommons.org/licenses/by-nc-nd/4.0/)
(http://creativecommons.org/licenses/by-nc-nd/4.0/)
Peer-review underresponsibility
Peer-review under responsibilityofof
thethe scientific
scientific committee
committee of the
of the 10th10th
CIRPCIRP Sponsored
Sponsored Conference
Conference on Digital
on Digital Enterprise
Enterprise Technologies
Technologies (DET
(DET
2020) –2020) – Digital
Digital Technologies
Technologies as Enablers
as Enablers of Industrial
of Industrial Competitiveness
Competitiveness and Sustainability.
and Sustainability.
Keywords: assembly line analysis, manufacturing management, discrete event-based modelling, mixed integer programming
𝑅𝑅𝑅𝑅𝑅𝑅𝑠𝑠𝑖𝑖,𝑗𝑗 ; 𝐹𝐹𝐹𝐹𝐹𝐹𝑙𝑙𝑖𝑖,𝑗𝑗 ; 𝑀𝑀𝑀𝑀𝑀𝑀𝑒𝑒𝑖𝑖,𝑗𝑗 ; 𝑆𝑆𝑆𝑆𝑆𝑆𝑑𝑑𝑖𝑖,𝑗𝑗 ; 𝐵𝐵𝐵𝐵𝐵𝐵𝑓𝑓𝑖𝑖,𝑗𝑗 =? (5) 1067 with the average cycle time, and 1432 units can be sold
keeping the fastest pace. With buffers the average negotiable
Where: number 1073 unit, and the peak yield is 1858 unit theoretically
j is the index of resource based on the collected data. These numbers match the numbers
derived from the statistically processed dataset. It means that
The cycle times calculated during the processing of the the model can be used in different scenarios as a reference for
dataset can be given as input. These cycle times will be the verification.
capacity consumption (cj) for every operation for each run. The It is interesting to observe the bottleneck. Considering the
total capacity consumption cannot be more than Ccap (6). average capacity consumption pro units without buffering, the
slowest operation is the 6th, while at a high production rate, the
𝐶𝐶𝑐𝑐𝑐𝑐𝑐𝑐 ≥ 𝑅𝑅𝑅𝑅𝑅𝑅𝑠𝑠𝑖𝑖,𝑗𝑗 ∙ 𝑐𝑐𝑗𝑗 , ∀𝑖𝑖, 𝑗𝑗 (6)
8th operation limits the throughput based on the stations’
As the nomenclature shows, there are buffers after each utilization. With buffers the theoretical limit is influenced only
resource and every forwarded unit was considered a sold unit. by the last operation – assuming there is enough time before a
These sales are internal ones, except at the end of the line. In high demand to fill the buffers appropriately. Of course, the
order to calculate real scenarios, certain integrity restrictions sold amount can be increased by stocking the product for later
have to be set. It is evident that the computed values of the sale when the demand is lesser than the available capacity, but
variables have to be non-negative integers (7). Naturally, the this results in additional holding costs and requires area in the
sold amount at the end of the line has to match the demand in storage. Since the usual demand is less than the maximal yield
the given time bucket (9). The demand for every internal sale capacity of the assembly line, keeping these units in the
can be calculated with a retrograde calculation minding the inventory is neither reasonable nor economical.
failed runs of the processes, too (8). An initial buffered quantity
(B0j) input is required to enable analyses from any time. It will 5. Discrete event-based simulation
mean an additional constraint in the program regarding the
zeroth period (10). A more detailed and subtle model was built to observe the
material flow on the whole assembly line. There are 26
𝑅𝑅𝑅𝑅𝑅𝑅𝑠𝑠𝑖𝑖,𝑗𝑗 ; 𝐹𝐹𝐹𝐹𝐹𝐹𝑙𝑙𝑖𝑖,𝑗𝑗 ; 𝑀𝑀𝑀𝑀𝑀𝑀𝑒𝑒𝑖𝑖,𝑗𝑗 ; 𝑆𝑆𝑆𝑆𝑆𝑆𝑑𝑑𝑖𝑖,𝑗𝑗 ; 𝐵𝐵𝐵𝐵𝐵𝐵𝑓𝑓𝑖𝑖,𝑗𝑗 ≥ 0 , ∀𝑖𝑖, 𝑗𝑗 (7) operations altogether in the four known cells. We used the
𝑆𝑆𝑆𝑆𝑆𝑆𝑑𝑑𝑖𝑖,𝑗𝑗−1 = 𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑖𝑖,𝑗𝑗 + 𝐹𝐹𝐹𝐹𝐹𝐹𝑙𝑙𝑖𝑖,𝑗𝑗 , 𝑗𝑗 ∈ [2; 𝑅𝑅], ∀𝑖𝑖 (8) Siemens Tecnomatix Plant Simulation software. The program
𝑆𝑆𝑆𝑆𝑆𝑆𝑑𝑑𝑖𝑖,𝑅𝑅 = 𝑑𝑑𝑖𝑖 , ∀𝑖𝑖 (9) offers a complex solution to simulate and optimize shop floors.
𝐵𝐵𝐵𝐵𝐵𝐵𝑓𝑓𝑖𝑖,𝑗𝑗 = 𝐵𝐵0𝑗𝑗 𝑖𝑖 = 0 , ∀𝑗𝑗 (10) Simple models can be built in this environment quickly with
the help of its graphical interface. Methods written in SimTalk
The process has no other source or sink except the listed can also be added in order to reach a more sophisticated, more
ones. This leads to further restrictions. These are as follows: refined result.
The number of total runs can be calculated by adding the Every cell was modelled in individual frames with dedicated
number of fail runs to the amount of yield individually for workers. Each operation was created as a station object. To
every resource in each period (11). The sold amount in a given ensure the material flow, the stations were connected with
period cannot exceed the sum of the yield in the period and the connector objects. In this model buffers were added only where
previously buffered amount (12). The buffered amount can be they existed in the factory, as well. Each object was placed like
calculated by increasing the previously buffered amount by the it was on the layout of the shop floor. Besides the labels of
difference of the yield and the number of forwarded units (13). objects, this also helped to identify the processes. The
The number of fail runs can be calculated by multiplying the individual frames were connected via interfaces. During the
runs by failure rate (Frj) (14). creation of the discrete event-based model, the actual buffers
were considered. These are the following: one buffer after
𝑅𝑅𝑅𝑅𝑅𝑅𝑠𝑠𝑖𝑖,𝑗𝑗 = 𝑀𝑀𝑀𝑀𝑀𝑀𝑒𝑒𝑖𝑖,𝑗𝑗 + 𝐹𝐹𝐹𝐹𝐹𝐹𝑙𝑙𝑖𝑖,𝑗𝑗 , ∀𝑖𝑖, 𝑗𝑗 (11) Cell1. There are two buffers within Cell2 and one after it to the
𝑆𝑆𝑆𝑆𝑆𝑆𝑑𝑑𝑖𝑖,𝑗𝑗 <= 𝑀𝑀𝑀𝑀𝑀𝑀𝑒𝑒𝑖𝑖,𝑗𝑗 + 𝐵𝐵𝐵𝐵𝐵𝐵𝑓𝑓𝑖𝑖−1,𝑗𝑗 , ∀𝑖𝑖, 𝑗𝑗 (12) Final Cell. There are two buffers to the Final Cell and one to
𝐵𝐵𝐵𝐵𝐵𝐵𝑓𝑓𝑖𝑖,𝑗𝑗 = 𝐵𝐵𝐵𝐵𝐵𝐵𝑓𝑓𝑖𝑖−1,𝑗𝑗 + 𝑀𝑀𝑀𝑀𝑀𝑀𝑒𝑒𝑖𝑖,𝑗𝑗 − 𝑆𝑆𝑆𝑆𝑆𝑆𝑑𝑑𝑖𝑖,𝑗𝑗 , ∀𝑖𝑖, 𝑗𝑗 (13) the External Cell after Cell2 and Cell3, each. The parts arriving
𝐹𝐹𝐹𝐹𝐹𝐹𝑙𝑙𝑖𝑖,𝑗𝑗 = 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑖𝑖,𝑗𝑗 ∙ (
𝐹𝐹𝑟𝑟𝑗𝑗
) , ∀𝑖𝑖, 𝑗𝑗 (14) from the External Cell are stored in two buffers before Cell2
100 and two before the Final Cell. There is a last buffer before the
The program needs a target function to find an optimal point last operation of the Final Cell.
in the search range. Since the main aim of this rough model is Cycle times, buffer capacity caps and other variables were
to verify the event-based model based on the numbers gained given in tables and global variables. These variables were based
from the statistical process, the objective function (15) is on the previously verified dataset. A method imported these
simple. In order to avoid unnecessary buffer levels, the total data from a table to every object based on the global variables.
number of buffered units was penalized. The weight of this This provided a quick and convenient way to change the
penalty is wb. necessary properties of the objects. It is a crucial aspect because
the number of operators in the cells influences the cycle time.
max(∑𝑇𝑇𝑖𝑖=1 𝑆𝑆𝑆𝑆𝑆𝑆𝑑𝑑𝑖𝑖,𝑅𝑅 − ∑𝑇𝑇𝑖𝑖=1 ∑𝑅𝑅𝑗𝑗=1 𝑤𝑤𝑏𝑏 ∙ 𝐵𝐵𝐵𝐵𝐵𝐵𝑓𝑓𝑖𝑖,𝑗𝑗 ) , 𝑤𝑤𝑏𝑏 > 1 (15) The problem with the described mathematical model is the
lack of non-deterministic effects. Although the statistical data
These equations are sufficient for basic checking. If no process integrated them into the results, the model itself is still
buffering is allowed, the maximal yield at the end of the line is strictly deterministic. The Plant Simulation software gives an
104 Norbert Ibriksz et al. / Procedia Manufacturing 54 (2021) 100–105
Author name / Procedia Manufacturing 00 (2019) 000–000 5
sufficient capacity for much longer than any MTTR in the Acknowledgements
system or any other planned downtime. This means that the size
of buffers is unnecessarily over capacitated. Work for this paper was partly supported by the European
Commission through the H2020 project EPIC under grant No.
6. Discussion 739592. This study was worked out in the frame of the research
project “Digital solutions for efficiency improvement of the
Building the mathematical model can be more challenging workers in production logistics” (registration number: 2017-
due to the required abstraction, but it is easy to extend further. 1.3.1-VKE-2017-00036). The results introduced in this paper
They can also be helpful when it is combined with artificial are applied in project No. ED_18-22018-0006, which project
intelligence. This time only basic calculations were executed was implemented with support provided by the National
with the help of it. Further extension is planned to determine Research, Development and Innovation Fund of Hungary, and
more economic buffer sizes. was financed in the scope of (publicly funded) funding schemes
The Plant Simulation model was deployed for a detailed according to Section 13(2) of the Hungarian Act on Scientific
analysis. We suggested to our partner, based on the results, to Research, Development and Innovation.
regroup workers from the Final cell in the case it is needed. The authors acknowledge that the final project of Krisztina
Furthermore, the unnecessarily big sizes of buffers were Szabó [6] was a great contribution to the establishment and
identified. analysis of the discrete event-based model.
One of the biggest obstacles was the lack of proper
information or its reliability during model construction. References
Although our partner has a MES, several kinds of information
could not have been queried from the database. On the other [1] Harrington Jr Dr J. Computer Integrated Manufacturing, New York:
hand, there is plenty of information in the MES that we did not Industrial Press Inc., 1973
[2] Nance RE, Sargent RG. Perspectives on the Evolution of Simulation. Op
need. This underlines the importance of proper data collection. Res 2002;50(1):161-172. https://doi.org/10.1287/opre.50.1.161.17790
Although modern technology offers several solutions to make [3] Szaller Á, Béres F, Piller É, Gyulai D, Pfeiffer A, Benczúr Ifj A. Real-
plants transparent, gathering all the data cannot be the ideal time prediction of manufacturing lead times in complex production
solution. It generates such an amount of raw data, unnecessarily environments In: 25th Annual EurOMA Conference. To Serve, To
increasing the investment costs by increasing data storage Produce and to Servitize in the Era of Networks, Big Data and Analytics;
Budapest, Hungary, 24 June-26 June 2018, pp. 1-10. European
capacity and communication speed requirements. With the Operations Management Association
increment of the attributes above, its price is usually growing [4] Li W, Huynh BH, Akhatar H, Myo KS. Discrete Event Simulation as a
exponentially, but at least proportionally. In addition, more Robust Supporting Tool for Smart Manufacturing In: Implementing
gathered data usually means more sensor, thus more Industry 4.0 (Toro C, Wang W, Akhatar H); 287-312; Cham, Springer
auxiliaries, as well. It also increases investment costs. This International Publishing, 2021
[5] Oesterle J, Bauernhansl T, Amodeo L. Hybrid Multi-objective
means that the purpose of the data collection has to be declared Optimization Method for Solving Simultaneously the line Balancing,
clearly before the implementation of the system, providing the Equipment and Buffer Sizing Problems for Hybrid Assembly Systems,
opportunity to determine the structure of the dataset carefully. Proc CIRP 2016;57:416-421
On the contrary, the more transparent is a plant, the more ISSN 2212-8271,https://doi.org/10.1016/j.procir.2016.11.072.
optimized performance can be achieved. Furthermore, we are [6] Ağpak K, Gökçen H. A Chance-Constrained Approach to Stochastic Line
Balancing Problem,Int J of Op Res 2007;180(3):1098-1115.
never able to tell whether we will need a specific information [7] Pınarbaşı M, Alakaş HM. Balancing stochastic type-II assembly lines:
in the future. In addition, the more sensor is placed, the more chance-constrained mixed integer and constraint programming models,
money can be saved in the future. All in all, the investment in Engineering Optimization 2020; 52(12): 2146-2163,
IoT systems is a definitely returning one. https://doi.org/10.1080/0305215X.2020.1716746
[8] IBM®. Mathematical programming versus constraint programming In:
IBM® Decision Optimization CPLEX® Modeling for Python
7. Summary (DOcplex) V2.20 documentation,
http://ibmdecisionoptimization.github.io/docplex-doc/mp_vs_cp.html
In this paper a mathematical and a discrete event-based (2016, accessesd: 04 May 2021)
model was presented. The input of the simulations was based [9] Lu K, Mizuno S, Shi J. A new mixed integer programming approach for
on a dataset collected in a real assembly system via a MES. The optimization over the efficient set of a multiobjective linear
programming problem, Optimization Letters 2020; 14:2323-2333,
limits of the system were calculated with the help of the https://doi.org/10.1007/s11590-020-01554-7
models. The effect of the number of operators was analyzed [10] Szabó K. Material-flow analysis on an electronical unit assembly line –
using the more detailed event-based simulation, providing master thesis, Budapest University of Technology and Economics, 2020
suggestions for worker reassignment. Buffer characteristics
were also examined. This showed that the system has oversized
buffers. Further analysis is planned to determine more
economic buffer sizes.