Flow Analysis and Optimization in Machining Lines - Analytical Models and Applications to a Real Case

Introduction
of different system configurations and set of parameters, in order to develop
the optimal configuration.
Research Problem and Motivation
The present master thesis work originates in the context of manufacturing
system performance analysis and optimization, with the aim to propose a
new analytical model for performance evaluation of two-machine lines char-
acterized by deterministic processing times, finite buffer capacity and ma-
chines affected by both time-dependent and operation-dependent failures.
It also presents an in-depth study of a multinational automotive company
manufacturing line with regard to the system analysis and performance im-
provement.
Starting from the industrial case study, we recognized the possibility to
develop a new analytical model for performance evaluation of transfer lines
submitted to particular features. Existing analytical models usually deal
only with operation-dependent failures1 because they are the majority in
most manufacturing systems and, at the same time, they simplify the model
mathematical treatment. On the other hand, most of modern transfer lines
are affected not only by such disruption types, but also by time-dependent
failures2. These disruptions are only influenced by how long a machine is
turned on and can therefore occur also if the machine is not producing a
1Operation Dependent Failures (ODFs) are failure types that occur only if the machine
is processing a part, for instance a tool breaking.
2Time Dependent Failures (TDFs) are failure types occurring whether or not a machine
is processing a part, such as electronic failures.
xxii
Introduction
part. One typical example is the spindle of a lathe that is frequently kept ro-
tating, even if no parts are being produced, yet the lathe may overheat (and
thus breakdown) any time the spindle is rotating. In literature the analytical
solution is presented only for lines affected either by different operation-
dependent disruptions or by one single time-dependent failure. We think it
is necessary to consider explicitly TDFs, if the aim is to derive more and
more accurate estimations. Following this idea we propose, besides a first
model where the analysis is carried out for a single time-dependent failure,
also a multiple failure mode model, where machines can fail in two differ-
ent disruption types, the first one operation-dependent and the second one
time-dependent. With this second approach it is possible to show how the
difference in throughput estimation changes, when the probability of the
time-dependent failure increases.
With regard to the industrial case study, the analysis, the modeling and
the improvement of a transfer line at Scania is presented. Such line pro-
duces 6-cylinder engine blocks by means of 23 stations, consisting in NC
machines and other stations including light assembling, washing and a final
quality control. Being the line also equipped with a semi-automatic data
collection system, a big amount of production data is available in the system
database, but such information are useless, because no organized analysis
tool is available at the company. In the present work we aim at showing the
potentiality of analytical models in evaluating the most important produc-
tion performances of this manufacturing system and at demonstrating how
they can also be used to suggest improvement directions. It is possible to
principally orient these directions toward either the technology process and
the intrinsic machine reliability or the process management. Specifically,
xxiii
Introduction
while in the former case the objective is to increase the system production
rate through, for instance, the use of more efficient and effective tools, equip-
ment and machine structural components, in the latter one the focus shifts
to a better utilization of the already available resources, that is a system
re-configuration. Following the given distinction, this part of our work deals
with the second approach. Starting from the data collection using the com-
pany database, the line is then modelled using a well-known analytical model.
Performances like throughput, average buffer levels, starvation and blocking
probabilities with the related causes have been accurately evaluated. In the
end buffer optimization, repair crew optimization and time to repair reduc-
tion are identified being the best option to achieve significant increments in
throughput.
Thesis Outline
The thesis consists of six chapters and two appendixes. The information
included in the chapters are enough for a full understanding of the work, in
the appendixes are reported details regarding the analysis tools and further
numerical results.
Chapter 1 provides a general introduction to manufacturing systems,
with particular attention to transfer lines, and to analytical models as perfor-
mance evaluation tools useful during the system design and running phases.
An effective comparison between analytical models and simulation tech-
niques, matched with specific literature references, is also proposed. In the
end we give a description of the main modeling assumptions shared by analy-
tical models for transfer lines.
In Chapter 2 a comprehensive literature overview regarding analytical
xxiv
Introduction
models is presented; our focus is principally on analytical models for flow line
performance evaluation and improvement, even if we refer also to models for
other types of manufacturing systems.
Chapter 3 and Chapter 4 are dedicated to the development of two new
models to address transfer lines characterized by deterministic processing
times, finite buffer capacity and time-dependent failures. In particular, in
Chapter 3 the single failure mode analysis is presented. Such analysis is
the starting point to develop, in Chapter 4, the solution for the multiple
failure modes case. In both chapters a particular attention to model as-
sumptions, terminology, conventions, as well as to solution technique ex-
planation is payed. In the end of these chapters and in Appendix A also
models consistency evaluation and comparisons with other existing models
are proposed.
In Chapter 5 we present the industrial case study. After a general
presentation of the multinational automotive company, we focus on the in-
depth description of the manufacturing system. In particular we give an
account for both the physical system (layout, machines, equipment, workers
and item produced) and the control system (the supervisor and the data
collection system).
Finally, Chapter 6 is dedicated to the modeling and the analysis of the
industrial case study. Our operative methodology is proposed clearly and step
by step, starting from the way we collected and analysed the shop floor data
and arriving to system performance evaluation and improvement solution
proposals. Causes mainly affecting the performances of the line are analyt-
ically identified and three different studies for improving the manufacturing
system are proposed and their increments in throughput are quantified.
xxv
Chapter 1
Manufacturing System Modeling
In this chapter a general introduction to manufacturing systems and to
the most common tools used to evaluated their performances is given. Section
1.1 provides a rapid description of different types of production systems and
the related features. In particular the focus is on transfer lines and their
typical behaviour. Section 1.2 compares the two most common tools used by
researchers and manufacturing engineers to model the production systems:
simulation and analytical models. In particular the attention is put on the
latter tool and on some related general modeling assumptions.
1.1 Manufacturing System Overview
A manufacturing process is defined the transformation of raw materials
and/or informations into finished goods and/or services. It also includes all
the intermediate processes involving the production of semi-finished parts.
There are three main kinds of input needed by a typical manufacturing pro-
cess: raw materials, manpower and production equipments. Raw materials
1
Chapter 1. Manufacturing System Modeling
Figure 1.1: Three-Axis Classification of Manufacturing Systems
are both materials and components, while production equipments are both
machines, support components and devices.
A general classification of industrial manufacturing systems is given by
A. Brandolese, G. Brugger, M. Garetti and M. Misul (1985) [1], where three
different classification axises are taken into account, as shown in figure 1.1.
The first one (the management axis) takes into account the different ways
to realize the production volume: in particular the production can be uni-
tary, submitted to batches or continuous. The second classification axis (the
market axis) consider the way to fulfil the demand: a company producing
on forecast manufactures and stores goods in warehouses because it started
producing before the orders took place. On the other hand, make to order
production starts only after a new order has come. With reference to this
second type, it can be further distinguished between production based on a
2
Chapter 1. Manufacturing System Modeling
single order and on repetitive orders. The third classification axis (the tech-
nological axis) focuses on the nature of the process: plants can be classified as
process plants or as production plants. In the formers deep chemical/phys-
ical transformations take place, thus there is no way to turn back, starting
from the the final good, to the original components. The latter includes both
machining (where single parts are realized) and assembly plants (where two
or more parts are put together to realize a finished good).
Production plants can be mainly organized in job-shops, manufacturing
cells, FMSs/FMCs and transfer lines.
In job-shops machines similar for technology stand close to each other in a
clearly defined area and every kind of product is characterized by one or more
alternative routings. Such type of systems are generally submitted to some
disadvantage, such as high levels of work in process (WIP), the difficulty in
following its flow, high cycle times, low machine saturations and high mix-
dependent performances. On the other hand, high flexibility of the mix and
the opportunity to perform alternative cycles are the main advantages.
In manufacturing cells production regards goods of the same family. Ma-
chines are grouped to produce only a specific family of products. Problems
related to this type of layout are, above all, the unbalancing of work contents
and the difficulty in managing periods with a turbulent mix. On the other
hand, the main strength aspects are the relatively short lead time, the low
WIP level, the high machining station saturations and the low number of
set-up.
In transfer lines, better analyzed in the following section, the semi-finished
goes from a machining station to another one thanks to a transfer system. A
solution like this usually warrants high production rates and efficiencies, but
3
Chapter 1. Manufacturing System Modeling
at the same time it bounds the flexibility in the meaning of the possibility
to work a wide family of parts.
FMS/FMCs can be considered, respectively, as a "flexibilization" of a
transfer line or an "automatization" of a manufacturing cell. They allow
a very flexible manufacturing, in order to face frequent mix changes, but
unfortunately they are often characterized by high capital investments.
1.1.1 Transfer Line Behaviour
A transfer line (also called flow line or production line) is a manufacturing
system, in which service stations or machines (M1, M2, ..., Mk), each other
separated by buffer storages (B1, B2, ...,Bk−1), are usually disposed in a linear
network [2]. Material flows, exactly once, from outside the system to M1,
then to B1, then toM2, and so forth until it reachesMk, after which it leaves
the system. Finite material storage areas distinguish flow lines from other
queuing systems. Work area may consists of a single machine or multiple
machines in parallel, in what S. B. Gershwin (1994) [2] calls series-parallel
systems.
Figure 1.2 shows a general transfer line. The squares represent the ma-
chines, while the circles represent the buffers. If machines behaviour was per-
fectly predictable and regular, there would be no need for buffers. However,
all the machines eventually fail, and some stations require an unpredictable,
or predictable, but variable, amount of time to complete their operations.
This unpredictability or abnormality has the potential for disrupting the
operations of the contiguous machines, or even machines further away, so the
buffers are used to reduce this eventuality.
Machines can be either up or down, that is productive or unproductive.
4
Chapter 1. Manufacturing System Modeling
Figure 1.2: General Transfer Line
When a failure occurs or when a machine takes an extraordinary long time to
complete an operation, the level in the adjacent upstream buffer may rise. If
the disruption persists long enough, that buffer fills up and forces the machine
upstream of it to stop processing parts. Such a forced state of down machine
is well called blocking. In the same way, the level of the adjacent downstream
buffer may fall during a failure, because the downstream machine drain its
contents; so if the failure persists long enough, the adjacent downstream
buffer empties and the machine downstream of it stops processing parts.
Such a forced state of down machine is, on the other hand, called starvation.
Blocking phenomenon propagates upstream the line, starting from the failed
machine and with a decreasing effect toward the first machine, while star-
vation propagates downstream the line, with a decreasing effect toward the
last machine. As a result the throughput of the line decreases.
By supplying both workpieces and room for workpieces, inter-operational
buffers partially de-couple adjacent machines. In fact, because of machine
failures are quite inevitable, since a faultless preventive maintenance can be
achieved only with a very high effort, the effects of machine failure on the
operations of the others is mitigated by the buffer storages. Nevertheless,
when buffers are empty or full this de-coupling effect cannot take place.
Consequently, as the buffer size increases, the probability that the buffer
being empty or full decreases and the effects of failures on the production
rate of the system are reduced, but more WIP parts are present between
5