International Association for Management of Technology
IAMOT 2010 Proceedings
MEASURING THE PERFORMANCE OF ASSEMBLY PROCESSES USING THROUGHPUT
CURVES
JOHANNES HINCKELDEYN*
Hamburg University of Applied Science
Department Mechanical Engineering and Production
Berliner Tor 21, 20099 Hamburg, Germany
Email: [email protected]
Dennis Kubera
Hamburg University of Applied Science
Department Mechanical Engineering and Production
Berliner Tor 21, 20099 Hamburg, Germany
Email: [email protected]
Nils Altfeld
Hamburg University of Applied Science
Department Mechanical Engineering and Production
Berliner Tor 21, 20099 Hamburg, Germany
Email: [email protected]
Jochen Kreutzfeldt
Hamburg University of Applied Science
Department Mechanical Engineering and Production
Berliner Tor 21, 20099 Hamburg, Germany
Email: [email protected]
*corresponding author
Abstract: The performance in many manufacturing companies is limited through bottlenecks in both production and assembly divisions. For this reason, bottlenecks are often the starting point of improvement initiatives. While there has been much
work done on bottlenecks that appear in production systems, bottlenecks in assembly processes are often overlooked. Due to
their location at the end of the manufacturing process however, bottlenecks in assembly processes have an impact on the
throughput of the entire system. To achieve optimal assembly performance, a coordinated and concerted consolidation of material flows into the assembly is vital. The presented work is completed as part of the research project "DePlaVis" that is located at the University of Applied Sciences in Hamburg. The starting point of its development is the throughput curve, which
identifies and evaluates bottlenecks in the manufacturing environment. To create the new assembly throughput curve, the
most influential factors affecting the assembly process are sought and two factors are identified – the workload and material
availability. The relationship between these two factors is mapped and a throughput curve was determined for an assembly
station. In order to verify the findings, an event-discrete simulation was programmed and analyzed. The simulation that is
conducted supported the developed model. It produces similar results for the throughput and thus for the performance. The
possible applications of the assembly throughput curve lies principally in the planning and control of orders for assembly
processes. Three planning cases and six operating cases have been identified. Each of these basic cases can be extracted and
displayed using the operating curve. Furthermore, it is also possible to derive information from the curve that can be used to
improve performance and efficiency.
Keywords: bottleneck management, throughput curve, assembly, performance measurement
Johannes Hinckeldeyn; Dennis Kubera; Nils Altfeld; Jochen Kreutzfeldt
Measuring the performance of assembly processes using throughput curves
1
International Association for Management of Technology
IAMOT 2010 Proceedings
Introduction
The overall performance of production processes is determined from the assembly performance. Nearly
all manufacturing processes involve some kind of assembly operation (Hopp & Spearman 2000, p. 315).
Assembly is defined as all activities, which are necessary to build products together from several parts
or components (VDI 1990, p.2). The assembly process is usually located in the end of the overall production process chain. Therefore the delivery performance is strongly depending on the performance of
the assembly process. For the assembly of products, all necessary parts and components have to be
available and sufficient capacity has to be provided for the operation. If only one factor is missing, the
respective order cannot be processed. The order decoupling point has also to be considered, because it
affects the point in the production where the customer requirements are taken into account (Dekkers
2006, p. 4014). In the case of engineer to order, make to order or assemble to order products, the customer decoupling point is located in advance to the assembly process and the availability of the right
material plays an important role. Additionally, variations and uncertainties in the upstream functions, for
example the delayed supply of components, can cause bigger variations in the downstream processes.
This phenomenon is known as the Bullwhip Effect (Lee et.al. 1997, p. 93). Bottlenecks in the upstream
process steps are threatening the outcome of the whole process and can constrain the organization from
fulfilling customer due dates and lowers the logistical performance of the whole production system
(Wiendahl 2002, p.3). Apart from mass production processes, assembly is cost intensive and cause
therefore up to 70 percent of the overall production costs (Lotter & Wiendahl 2006, p.6). Assembly
processes are usually costly, because of the high investment volume into automation (Boysen et.al.
2007, p. 676). In an increasing high-tech world, robotic systems offer good perspectives for the automation of assembly activities. Rampersad (1995) discusses some of the problems that thwart the widespread development and application of such systems. However, improvement measures have to be undertaken by assembly workers. For the implementation of these measures, the shop floor workers have
to be convinced and inspired. Visual approaches have a good acceptance in practice (Eppler & Mengis
2009, p.50) and a graphical solution for linking capacity and material availability data to assembly performance and planning quality information is preferred.
Research Project DePlaVis – Development of a visual bottleneck approach
The research project DePlaVis was introduced to develop a visual bottleneck management solution for
production systems. The project is founded by the German Federal Ministry of Education and Research.
The consortium of the project consists of three mechanical engineering companies, two software houses
and two Universities. The bottleneck management approach of DePlaVis is based upon the theory of logistic operation curves (Nyhuis & Wiendahl 2003). In contrast to the original approach, the research
team uses the throughput curve, which is able to visualize and assess bottlenecks in the production system (Schultheiss & Kreutzfeldt 2009). This solution draws the causal relation of particular bottlenecks
and the overall system performance. With the developed software demonstrator on hand, the throughput
curve can be computed and directly communicated on the shop floor to discuss countermeasure or improvement initiatives. The investigation in the three engineering companies underpinned the need for a
bottleneck management solution, which is easy to handle and understand in practice. Today the
throughput curve is restricted to capacity bottlenecks. However the material supply plays a big role in
assembly processes and the availability has to be considered, when bottlenecks are computed and evaluated.
Objective and structure of the paper
For the application of throughput curves in assembly processes, it is therefore important to integrate the
material availability into the bottleneck management concept. The objective of this paper is to demonstrate and verify a possible visualization for bottleneck management in assembly processes to underJohannes Hinckeldeyn; Dennis Kubera; Nils Altfeld; Jochen Kreutzfeldt
Measuring the performance of assembly processes using throughput curves
2
International Association for Management of Technology
IAMOT 2010 Proceedings
stand and communicate bottleneck problems on the shop floor. The paper is structured as follows: In the
next chapter, planning and control approaches of assembly processes under consideration of the theory
of logistic operating curves are discussed. Thereafter, the approach for bottleneck management with
throughput curves is outlined. Based on the appearing gaps a possible solution is developed with differentiation of nine original cases of planning and operation. This throughput curve is verified and discussed with an event-discrete simulation model. At the end of the paper, possible areas for further research and next steps are identified for the planning and control bottlenecks in assembly processes.
Planning and Control of Assembly Processes
Assembly planning and control is a well addressed topic in literature. Addressed topics are planning and
scheduling approaches, bioanalogical solutions, heuristics, mathematical models and assembly line balancing. The last issue seeks to cover the problem, how to configure the planning of assembly lines. It
can be divided into Simple Assembly Line Balancing (SALB) and General Assembly Line Balancing
(GALB) (Boysen et.al. 2007, p. 675). SALB comprise assigning task to assembly lines under simplified
assumptions, whereas GALP tries to integrate more problem parameters, like the assembly line layout.
Another problem to assembly line balancing is the consideration of hybrid flow shops, which has a heterogeneous structure in the layout of the respective assembly lines (Ribas et al 2010, p.1439).
Many of the assembly planning and scheduling approaches are bottleneck based and several solutions
for dispatching rules are developed. For example Salegna and Park (1994, p.91) suggest a solution,
which considers load smoothing in the order release process. Further Rajendran and Alicke (2007, p.89)
have developed a set of dispatching rules for static flow shops with bottleneck machines. They consider
this problem using three measures of performance - minimization of total flow time of jobs, the minimization of the sum of earliness and tardiness of jobs, and the minimization of total tardiness of jobs. The
findings of their experimental investigation have been measured against conventional dispatching rules.
Another way for planning and scheduling is the application of bioanalogical approaches, like insect algorithms from ants and wasps (Cicirello & Smith 2001, p.1; Cicirello & Smith 2004, p.237). The advantage of these solutions is the robustness against unpredictable and dynamic changes. Song et al. (2007,
p.569) expand biological algorithms on the field of optimization called ant colony optimization (ACO).
The authors formulate the bottleneck station scheduling problem, and then apply ant colony optimization (ACO) to solve it metaheuristically.
Heuristics are an often preferred solution for bottleneck-based planning and scheduling in assembly
processes. Chen and Chen (2009) have developed a bottleneck based heuristic to solve flexible flow line
problems that contain a bottleneck stage. The objective of the heuristic is to minimize makespan. Computational results show that it significantly outperforms all the well-known heuristics. Furthermore, this
method can also be applied to solve other scheduling problems such as job shop problems with bottleneck stages. An approach, which expands heuristic to the whole supply chains is proposed by Kung and
Chern (2009, p.2513). It is called the Heuristic Factory Planning Algorithm (HFPA). After identifying
the bottleneck work centre, the HFPA sorts the jobs using a series of criteria before the final planning
step. One method to improve planning heuristics are Petri Nets (Chauvet et al. 2003, p.150). The application of such enhanced heuristics has been simulated under restricted conditions with positive results.
For more exact solution, mathematical models for scheduling and control of assembly processes are developed. Aggarwal et al (1986, p.11) have developed a mathematical method for solving assembly line
problems for jobs that are dependant. The paper considers two cases for reducing the makespan. Manufacturing equipment on the factory floor is typically unreliable and the buffers are finite. Therefore production systems are stochastic and nonlinear. Ching et al. (2008, p.911) have developed a practical maJohannes Hinckeldeyn; Dennis Kubera; Nils Altfeld; Jochen Kreutzfeldt
Measuring the performance of assembly processes using throughput curves
3
International Association for Management of Technology
IAMOT 2010 Proceedings
thematical method for analysis and improvement of assembly systems with non-exponential machines.
The bottleneck is identified based on the frequencies of machine blockages and starvations.
In the discussed literature many bottleneck management approaches for assembly processes can be
found. However none of the presented solutions is able to visualize the interrelation between input parameters of an assembly line, the respective bottleneck station and the throughput of the system. The
development of an easily understandable tool would be very helpful to enhance the communication of
problems and countermeasures on the shop floor.
Bottleneck Management with Throughput Curves
The first approach to describe logistical systems also graphically was the funnel model, (Bechte 1984;
Wiendahl 1992). This analogy describes a work system, like a work station or even a whole factory. The
fill level of the funnel depicts the work load of the system. The width of the funnel opening, through
which the production orders flow, equates to the capacity. The relationship between these parameters
can be found of the throughput diagram, the first visual impression of the logistical parameters of a
work system. Input and output are depicted over time. Based on this analogy, the Theory of Logistic
Operating Curves was developed (Nyhuis 1991, 2006, 2007) (Wiendahl and Nyhuis 2003). In a deductive-empirical approach, the performance and the throughput time of a work system are considered as
dependent variables, which are related to the order backlog as independent variable. A C-Norm function
provides the mathematical basis for the computation of the logistical curves. As this approach is working with average values, the curve is only suitable for static analysis of logistical system. The application of logistic operation lies therefore more in the strategic positioning of logistical systems and not in
the operative control. Wiendahl and Hegenscheidt (2002) use an operating curve to describe the utilization of assembly lines as a function of the operational availability. Therewith the buffer sizes between
the individual workstations can be determined. However the curve for assembly processes operates also
with average values and its application lies more in the strategic area for decisions in assembly lines, especially in the dimensioning of buffer sizes between assembly stations. An operative bottleneck management is not considered within this approach.
The throughput curve as a bottleneck management instrument is developed by Kreutzfeld (1995). It
enables a bottleneck management with a visual approach linked to an approximation algorithm. To
represent the relationship between the input variables (e.g. load, work plans and capacity) and output variables (e.g. performance, utilization and inventory) of work systems, the throughput curve was selected
as the basis for the identification and evaluation of bottlenecks. To apply it, three assumptions in an
analogy to electrical systems are made (Kreutzfeldt 2007):
1. A bottleneck restricts the flow of production orders in a similar manner that a resistor limits the flow
of electrical current in an electrical network. This similarity can be assumed, if discrete order flows
are considered as continuous flows. The probability that an order is processed depends on the ratio
of the workload to the capacity at the work station with the greatest workload. This work station is
termed the throughput limiter. The throughput limiter is defined as the work station with the greatest
ratio of work load to capacity.
2. Thus, a parallel can be drawn between the continuous flow of orders through a production system
and an electrical current as it flows through an electrical network. All orders that move through the
same limiter become a continuous flow of orders. This flow contains the workload of all orders on
all work stations in a period.
3. In this way, just as an electrical network can be described by electric currents and resistors, a production network can be modeled based on bottlenecks and flows of production orders.
Johannes Hinckeldeyn; Dennis Kubera; Nils Altfeld; Jochen Kreutzfeldt
Measuring the performance of assembly processes using throughput curves
4
International Association for Management of Technology
IAMOT 2010 Proceedings
To better understand the calculation of the throughput curve, an example for a throughput curve is
drawn in Figure 1. The pictured throughput curve represents the performance of a single order stream.
The order stream is determined by the bottleneck work station. All orders, which go through this work
station, are summarized as one order stream. This follows the principle of Goldratt (1994), who states
that the bottleneck work station controls the performance of a system. In this example, the workload of
all orders within the order stream is summarized up to a work content of 120 hours. Due to the capacity
constraint of the bottleneck, the order stream is only able to transform 60 hours of the workload into
throughput and is shown as an angle bisector in the first half of the throughput curve. The remaining
workload of 60 hours builds inventory in front of the bottleneck and is called throughput potential. The
visual representation is a horizontal line in the second half of the throughput curve. The effect of the
bottleneck onto other work stations within the order stream can be estimated by differentiation of the
throughput into direct and indirect throughput. The overall throughput of the whole order stream is 60
hours. The amount of work processed on work station in front or behind the bottleneck is drawn as a
vertical line on the end of the throughput curve and called indirect throughput. The indirect throughput
in this example is 40 hours. The direct throughput is shown as a free area between the bottom of the diagram and the beginning indirect throughput line. It equals the work processed directly on the bottleneck
workstation. The relation of direct and indirect throughput is very important to assess the effect of bottlenecks on other workstations within the order stream. As bigger the amount of indirect throughput to
direct throughput as bigger is the effect of the bottleneck onto other work stations.
Figure 1: Example of a throughput curve according to Kreutzfeldt (1995).
The diagram shows visually two bottleneck parameters. First, the throughput potential shows the effect
of a bottleneck on the inventory within an order stream. In case of an optimization measure to decrease
inventory within the system, the bottlenecks with a long horizontal line should be addressed. Second,
the proportion of indirect and direct throughput indicates the effect of the bottleneck on other workstation. This means as a converse argument, that a capacity enlargement at this station promises a large effect of the overall system performance. In case of a productivity optimization of the whole system, the
bottlenecks with the biggest proportion of indirect and direct throughput should be addressed first.
This approach provides and easy to understand and communicate solution for bottleneck problems in the
manufacturing environment. However there are only capacity bottlenecks considered at the moment.
This a restriction for assembly processes, where the availability of material plays an important role as
well as the availability of capacity. To show the optimization potential in assembly processes the
Johannes Hinckeldeyn; Dennis Kubera; Nils Altfeld; Jochen Kreutzfeldt
Measuring the performance of assembly processes using throughput curves
5
International Association for Management of Technology
IAMOT 2010 Proceedings
throughput curve should be enhanced with a material dimension to use this visualization and communication tool in this part of a production system as well. The evaluation of bottlenecks gives the opportunity to prioritize bottlenecks regarding their optimization potential. This is very important if resources for
the bottleneck relief are scarce. The communication tool shows quickly this optimization potential and
so it is possible to take countermeasures, where they unfold the largest effect. This is important for
companies to ensure a sustainable bottleneck management.
´
Throughput Curves for Assembly Processes
The application of throughput curves in the manufacturing environment (Schultheiss & Kreutzfeldt
2009) marks the starting point for an expansion to assembly processes, which are particularly interesting
for bottleneck investigation. The assembly process, as the next step in the value chain, is depending on
the supply with parts and components, as well as the availability of capacity. The assembly is mostly the
last step in the production process. Problems in the manufacturing process have therefore a direct impact
on the performance of assembly processes, if required materials are not available. Delays from previous
processes are summarizing to an overall scheduling delay of the order until the last required component
has arrived. This is called the Forrester effect (Lee et.al. 1997) and usually responsible for further problems in the assembly process. An early prediction of these arising problems in manufacturing and their
effect on the assembly would make it possible to take well-timed countermeasures. Examples are capacity shortages or rising inventories. The objective to adopt throughput curves for assembly processes is to
identify and visualize bottlenecks, which are the basis for improvement measures in manufacturing and
assembly. Figure 2 shows the system, which has to be realized for control and permanent improvement
of assembly processes.
production
master data
bottleneck
management
DePlaVis
Material
availability
evaluated
bottleneck
visualisation
assembly
planning system
process
parameters
workload
parameters
supply chain
process
assembly process
throughput
Legend 2:
Material Flow
Information Flow
Figure 2: System architecture of an optimized assembly process.
The bottleneck management software DePlaVis is able to identify and evaluate bottlenecks based upon
planned workload and production master data. The relationship between the input of an assembly
process and the expected performance is drawn in an operating curve, so called throughput curve. This
graphic allows the assembly planer to take all necessary countermeasures to break the bottlenecks and
increase the assembly performance with optimized planning parameters. However the throughput curves
are currently only dealing with capacity bottlenecks. For a successful application of throughput curves
to assembly processes, the availability of material has to be integrated as an additional input factor.
Johannes Hinckeldeyn; Dennis Kubera; Nils Altfeld; Jochen Kreutzfeldt
Measuring the performance of assembly processes using throughput curves
6
International Association for Management of Technology
IAMOT 2010 Proceedings
Adaptation of a bottleneck management solution for assembly processes
According to Lotter and Wiendahl (2006, p.370) the performance of assembly processes is depending
on four areas of influence:
Organization
work stations
parts and components
system
An analysis of available datasets in existing ERP systems1 shows that the four areas of influence can be
linked to the datasets within the system. These datasets are further linked to the input parameters of the
throughput curve to ensure an automated processing. The relationship between areas of influence, ERP
datasets and input parameters of the throughput curve are shown in table 1:
Table 1: Relationship of input parameters and areas of influence in assembly.
Area of influence
ERP data sets (examples)
Input Parameters for Assembly Throughput Curves
Work station
Workload
Workload
Cycle time
Machine breakdown rate
Production master data
Organization
Capacity
Assembly operations
System
Assembly structure
Assembly elasticity
Parts and components
Availability
Material
Quality
The workload in a certain period has impact on the performance of an assembly process. The assembly
processes have to provide sufficient free capacity available to perform the assembly order. The amount
of free capacity in a certain period is computed from the shift plan of the company. In the case of machine use, the machine breakdown rate also has to be considered. The capacity is computed for single
workstations and aggregated for larger planning purposes, for example whole factories. The first parameter to be considered is the workload (W) for processing assembly orders. The workload is usually
computed under consideration of work plans, planned cycle times and amount of products to be assembled. To process the planned workload, the work system has to provide sufficient free capacity. The relationship of capacity and workload can be described in three cases:
1. sufficient capacity (capacity = workload)
2. excess capacity (capacity > workload)
3. insufficient capacity (capacity < workload)
Case 1. and case 2. are uncritical for assembly planning, because all planned orders can be processed
from a capacity perspective. Insufficient capacity (case 3) however leads to a capacity bottleneck and
increasing stock. Such a bottleneck can be hold responsible for tardiness of customer orders.
A further important input parameter is the availability of material and components, which are assembled
to fulfill the customer order (Jünemann et.al. 1989). Every part must be available in the respective
1
The applied ERP system in the research project DePlaVis was PSIPENTA (www.psipenta.de).
Johannes Hinckeldeyn; Dennis Kubera; Nils Altfeld; Jochen Kreutzfeldt
Measuring the performance of assembly processes using throughput curves
7
International Association for Management of Technology
IAMOT 2010 Proceedings
amount and quality to perform the assembly operation. The operation can be made impossible just by
the absence of a single component and can therefore lead to a delay of the whole customer order with effects to all following process steps. For that reason, the input parameter material availability (MA) is introduced to incorporate the material dimension in the assembly throughput curve.
The planned assembly throughput marks the starting point for the computation of MA. This parameter is
derived from the respective bill of material of the planned customer orders. The amount of material for
each component is necessary to process all planned orders within a period and is called material demand
MDx. For the computation of MA, it is important to distinguish in ERP datasets between self-made
components, which are manufactured by the company itself, subcontracted components, which are manufactured by outside suppliers, and components on stock, which are stored in the inventory. Availability
of self-made components can be estimated with the original throughput curve for manufacturing
processes (Schultheiss & Kreutzfeldt 2009). The availability of subcontracted components and components in stock are found in the datasets of the ERP system. Based on this raw data, the value of MAx is
computed by comparison of MDx with the expected material supply MSx of the component:
[1]
Hence, MAx is non-dimensional and shows the relative value of the demanded parts and components,
expected to be available. Three states of MA can be distinguished:
1. material sufficiency (MA = 1)
2. material shortage (MA < 1)
3. material oversupply (MA > 1)
Like in the capacity situation, case 1. and 3. are uncritical for the processing of assembly orders. But it
must be mentioned that a material oversupply can cause inefficiencies, for example higher storage costs.
A shortage of material however threatens the fulfillment of customer requirements. The new throughput
curve is developed based on the two dimensions workload W and material availability MA. An example
of the assembly throughput curve is shown in figure 3. To achieve a clear and easily understandable diagram, the abscissa standard mMA of the diagram should be equaled to the ordinate standard, where mW
counts for one unit:
mMA = mW * Cmax [2]
This standardization ensures an ideal assembly throughput curve with tan(φ) = 1 or φ = 45°.
Johannes Hinckeldeyn; Dennis Kubera; Nils Altfeld; Jochen Kreutzfeldt
Measuring the performance of assembly processes using throughput curves
8
International Association for Management of Technology
IAMOT 2010 Proceedings
W
Cmax
φ
1
MA
Legend 3:
estimated operating point under consideration of system constraints
C
capacity [hours]
MA
material availibility [dimensionless]
W
workload [hours]
Figure 3: Assembly throughput curve with operating point.
The x-axis shows MA as the independent variable, because the material availability is related to the performance of the manufacturing process of the company itself and of supplying subcontractors. The point
MA = 1 is particularly marked, as there is every material available to planned process the customer orders. This means in addition, that the material supply meets the material demand (MD x=MSx). The parameter workload (W) is shown on the y-axis. The maximum capacity Cmax has to be here particularly
marked, because exceeding workload will lead to a capacity bottleneck. Therefore, the optimal operating
point of the assembly system is the intersection point of MA =1 and W = Cmax and marked as a cross.
This point indicates the highest possible performance from the assembly system with consideration of
all system constraints. In this example all orders can be processed, because there is enough capacity
available and the material supply is sufficient. Additionally, no idle capacity or material oversupply is
wasted and the system works in the most efficient state. Based on this operation point, the throughput
curve can be drawn. As MA is the independent variable, a decrease will subsequently lead to a lower required capacity. On the other hand, an increase of MA (more material available as necessary) will have
no effect on the assembly performance, since more orders cannot be processed due to capacity restrictions. The assembly process and the respective manufacturing and supply chain process is optimal balanced. Hence, the assembly throughput curve can also assess the quality of planning between the assembly and the supplying processes. Nine possible cases and respective assembly throughput curves can
be derived based on the three situations for each state of MA and W. These nine cases are displayed in
the following figure 4.
Johannes Hinckeldeyn; Dennis Kubera; Nils Altfeld; Jochen Kreutzfeldt
Measuring the performance of assembly processes using throughput curves
9
International Association for Management of Technology
IAMOT 2010 Proceedings
W
MA
= Cmax
< Cmax
W
1
Cmax
> Cmax
W
2
W
CC
3
Cmax
Cmax
IC
=1
MA
1
W
4
Cmax
MA
1
W
5
Cmax
W
CC
6
Cmax
<1
MC
IC
MC
MA
1
MC
MA
1
W
7
MA
1
W
8
MA
1
W
MO
9
CC
Cmax
Cmax
Cmax
>1
IC
MO
MO
1
MA
1
MA
1
MA
Legend 4:
MO
MC
IC
CC
operating curve
estimated operating point under consideration of system constraints
material over supply
material constraint
idle capacity
capacity constraint
Figure 4: Nine possible cases for assembly throughput curves.
The cases 1, 2 and 3 are typical for planning. In this stadium, the planner assumes the full availability of
the activity-scheduled parts and components. Parameter MA contains therefore the value 1. This meets
the assumption that the demand of material can be fully satisfied by inventory, self-manufacturing or
subcontractor. The planned workload Wplan is derived from the planned amount of goods to be assembled multiplied with the process time per unit according to the working plan. The value of the workload
differs from the parameter Cmax, if the planned capacity is not well synchronized with the material
supply. The resulting point is marked with a cross. Case 2 lefts idle capacity (IC), as there is not enough
planned workload for the capacity supply. Case 3 on the other hand shows a capacity constraint (CC),
because the planned workload exceeds the capacity supply. Set-up time is neglected in all cases, but can
be easily added and would lead to a vertical adjustment of the curve according to the value of the set-up
time. Following improvement measures could be recommended:
-
Case 1: Optimal planning. No action is required.
Case 2: There is idle capacity created on the assembly station. It would be possible to reduce the capacity to streamline the assembly process or to dispatch more orders.
Johannes Hinckeldeyn; Dennis Kubera; Nils Altfeld; Jochen Kreutzfeldt
Measuring the performance of assembly processes using throughput curves
10
International Association for Management of Technology
IAMOT 2010 Proceedings
-
Case 3: A capacity bottleneck is going to appear. The capacity supply should be increased to process
all the assembly orders or the material availability can be reduced to avoid the build of inventory.
For the cases 4 to 9, the real material availability has to be computed. It is especially important to consider the material constraint, if necessary. That means the material with the least availability in the bill
of materials for the respective assembly must be identified. Therefore the material availability of every
component MAx is computed. Afterwards the minimum MAx is selected, which represents the material
shortage and therefore the assembly bottleneck. This value can be equalized to MA on the x-axis and is
the basis for further computation steps of the assembly throughput curve.
The cases 4 and 5 are only restricted by the material constraints (MC) and the capacity demand does not
exceed the capacity supply. The function to calculate the operating point is therefore only depending on
MA. The operating point shows additionally some idle capacity (IC), because of the undersupply with
material. Measures for performance improvement for the several cases could be as follows:
-
Case 4: An increase of material availability makes no sense without a synchronized increase of capacity, due to fully loaded capacity.
Case 5: An increase of material availability could be useful, because there is idle capacity left to
process more orders.
The cases 7 and 8 however are not restricted by any parameter. Both cases show a material oversupply
(MO) compared to the planned scenario. Due to the operating point below the maximum capacity, it is
possible to process more orders than required. Measures for performance improvement for the several
cases could be:
-
Case 7: This case has a planning issue, because there is more material available than for planned orders, but with enough capacity. It should be checked, if all assembled orders can be sold or used.
Case 8: This case can be compared to case 7, but with idle capacity. It would be possible to plan
more orders for assembly processing. Anyhow there is also a planning issue.
In the next step the capacity supply must also be considered. Despite the constraint of material availability, the capacity can limit the assembly performance in addition. It is therefore the question, what the
real bottleneck is. A capacity bottleneck is created, if the workload exceeds the capacity supply. This
happens in case 6 and 9. There might a material restriction anyhow, like in case 6. Nevertheless the operating point must be corrected twice in these cases and the function of the operating point is depending
on MA and Cmax. Measures for performance improvement for case 6 and 9 could be as follows:
-
Case 6: The capacity bottleneck restricts the assembly from processing all the planned orders, as
well as the material constraint. It has to be investigated, what the dominating constraint is and then
careful countermeasures can be undertaken.
Case 9: Capacity demand and material availability exceed the system capacities. Nevertheless, the
capacity shortage restricts the assembly from achieving more throughput. A capacity increase would
help to process more assembly orders.
The assembly throughput curves are able to visualize bottleneck parameters from planning and reality.
The planning and scheduling of assembly processes can be supported as well as the indication of optimization potential. Especially the quality of planning can be assessed with the curve. The solution will
be verified in the next section.
Johannes Hinckeldeyn; Dennis Kubera; Nils Altfeld; Jochen Kreutzfeldt
Measuring the performance of assembly processes using throughput curves
11
International Association for Management of Technology
IAMOT 2010 Proceedings
Verification through an event-discrete simulation model
The assembly throughput curve is verified with an event-discrete simulation model in two simulation
scenarios. The simulation model contains three work systems (WSi). These include two manufacturing
systems (WS1 and WS2), one assembly system (WS3). Furthermore there are two sources and one sink.
The sources are raw material suppliers to the manufacturing systems and supply continuously all raw
material demanded. The two manufacturing systems produce the two different parts P1 and P2. Both
parts are matched in the assembly system into product P3. It is important to notice, that one unit of P1
and two units of P2 are needed to assemble one unit of P3. A practical example could be the assembly of
two different gearwheels and one shaft. This ratio between P1 and P2 is called joining relationship and
can be found in the bill of material. The considered period is 60 minutes which equals also the maximum capacity Cmax of every manufacturing system and the assembly system.
Table 2: Simulation parameter for scenario 1.
i
WSi
Work system parameters
Work plan parameters
Description
Capacity Cmax
[min]
Product Pi
Joining relationship
[P1 : P2]
Cycle time t(Pi)
[min/unit]
1
WS1
Manufacturing 1
60
P1
n/a
0.25
2
WS2
Manufacturing 2
60
P2
n/a
0.25
3
WS3
Assembler
60
P3
1:2
2
First, it is assumed to manufacture one unit of P1 on WS1 and P2 on WS2 within a period of 0.25 minutes
each. The assembly process on WS3 takes 2 minutes per unit of P3. It is further assumed to manufacture
only one kind of component within the considered period on each manufacturing system. The planned
workload is computed as multiplication of planned production volume PVi and the respective cycle time
t(Pi):
Wplan(WSi) = PVplan(Pi) * t(Pi) [5]
The planned production volume PVplan(Pi), the computed workload Wplan(WSi) and the outcomes of scenario 1 can be found in table 3:
Table 3: Production plan and output of simulation model for scenario 1.
Product
Pi
Production plan
PVplan [units]
Output
Wplan [min]
[units]
[min]
P1
120
30
120
30
P2
240
60
239
59,75
P3
120
240
29
58
It is evident, that the manufacturing systems are nearly able to fulfill the production plan as given. Material is sufficiently available. Therefore parameter MA can be rounded to 1. The system constrain in
this process is the capacity of the assembly system (WS3), which can only provide a maximum capacity
(Cmax) of 60 minutes. This is nearly 25 percent of the demanded capacity of the assembly process. The
resulting assembly throughput curve of WS3 corresponds to case 3. Figure 5 shows the adjusted assemJohannes Hinckeldeyn; Dennis Kubera; Nils Altfeld; Jochen Kreutzfeldt
Measuring the performance of assembly processes using throughput curves
12
International Association for Management of Technology
IAMOT 2010 Proceedings
bly throughput curve. It is important to notice, that for displaying reasons, the upper part of the diagram
is abbreviated.
W
240
CC
60
0,24
1
MA
Legend 5:
CC
operating curve
estimated operating point under consideration of system constraints
capacity constraint
Figure 5: Assembly throughput curve of scenario 1.
The objective of the simulation is to investigate the influence of the material availability on the assembly system performance. Therefore parameter MA is successively reduced by increasing the cycle time
t(P1) from 15 seconds to 130 seconds. All other parameters of the simulation model remained constantly. Table 4 displays the new parameters for scenario 2 of the simulation:
Table 4: Simulation parameter for scenario 2.
i
WSi
Work system parameters
Work plan parameters
Description
Capacity Cmax
[min]
Product Pi
Joining relationship
[P1 : P2]
Cycle time t(Pi)
[min/unit]
1
WS1
Manufacturing 1
60
P1
n/a
2.17
2
WS2
Manufacturing 2
60
P2
n/a
0.25
3
WS3
Assembler
60
P3
1:2
2
Under the different conditions of scenario 2, the production plan changed as well. The new production
plan and the outcomes are shown in table 5:
Johannes Hinckeldeyn; Dennis Kubera; Nils Altfeld; Jochen Kreutzfeldt
Measuring the performance of assembly processes using throughput curves
13
International Association for Management of Technology
IAMOT 2010 Proceedings
Table 5: Production plan and output of simulation model for scenario 1.
Product
Pi
Production plan
PVplan [units]
Output
Wplan [min]
[units]
[min]
P1
120
269
27
58.5
P2
240
60
239
59,75
P3
120
240
26
52
The outcomes show that, despite constant capacity and cycle time of WS3, the throughput of the assembly process decreases. The bottleneck shifts from the assembly capacity to the availability of component
P1. This can be traced back to the capacity of WS1, which is not sufficient to process the necessary
amount of P1 to ensure full load of the assembly process. The respective assembly throughput curve in
figure 6 shows an operating point below the maximum capacity of Cmax(WS3) and represents case 5. The
system constraint is the availability of component P1 with the lowest value of MA (MA (P1) 0.22).
W
240
60
IC
52
MC
0,22
1
MA
Legend 6:
operating curve
estimated operating point under consideration of system constraints
MC
material constraint
IC
idle capacity
Figure 6: Assembly throughput curve of scenario 2.
The utilized capacity, which is needed for assembling all components of P1, is below the maximum capacity Cmax of the assembly process. It is therefore proved, that the bottleneck has shifted from the assembly capacity to the availability of P1.
Johannes Hinckeldeyn; Dennis Kubera; Nils Altfeld; Jochen Kreutzfeldt
Measuring the performance of assembly processes using throughput curves
14
International Association for Management of Technology
IAMOT 2010 Proceedings
Discussion
The outcomes of the simulation show in comparison an identical behavior to the proposed assembly
throughput curve. The parameters are easily extracted from the diagram and can directly used for communication. It is therefore easy to discuss and plan improvement measures on the shop floor and with
the production manager. A possible solution could be, not to release as much manufacturing orders as
possible to achieve a high utilization of the manufacturing system. As stated from Goldratt (1990), the
system has to be aligned on the overall system constraint, no matter if manufacturing or assembly. Such
a measure would prevent this system before generating inventory within its production process, which
would subsequently lead to longer and varied lead times and decreasing throughput.
However, this approach has currently several limits, which lead to further research. The described assembly throughput curve deals with stochastic probabilities and relies therefore on sufficient large numbers of parts and orders. This is mostly the case in serial or mass production and in contrast to make to
order or engineer to order products, which are usually unique. These production processes have to deal
with small numbers or even event discrete factors. This is especially important for the assembly, where
the availability of certain components has an enormous impact on the successful completion of an assembly order.
A further limit is the lacking integration of time in the computation of the assembly throughput curve.
The moment of arrival of certain parts is currently not considered. This plays an inferior role in the
analysis of serial production process. However, this is different to make to order or engineer to order
production processes, where the temporary lack of a certain component can hinder the progress of the
whole assembly. Further it is not possible to analyse different periods at the moment. It is therefore not
possible to identify dynamic bottleneck behavior, like shifting bottlenecks, without the repeated computation and manual comparison of different diagrams.
Concluding remarks
The assembly throughput curve is a useful tool for the identification of assembly bottlenecks and the respective optimization potentials. It is possible to evaluate the quality of assembly and production planning against the estimated situation in the assembly. This offers production managers the opportunity to
take countermeasures against bottlenecks. A first possible improvement activity could be the alignment
of assembly planning. It is also worth to think about activities for manufacturing or purchasing to ensure
a material supply according to the expected demand in the assembly process. All these parameters can
be easily visualized and discussed in the diagram of an assembly throughput curve. This approach is also a good communication tool to transfer identified optimization potential directly onto the shop floor
without the analysis of data and transformation into understandable Power Point slides.
All the described research work is carried out within the research project DePlaVis. The next steps are
the investigation of communication behavior and how effective this tool is applied in practice and the
estimation of the respective training effort to apply this visualization tool. It is therefore planned to test
the assembly throughput curve in three different companies with a distinctive assembly process.
As these companies are working in the engineer to order environment, the boundaries of the approach
have also to be investigated. It is likely, that the assembly throughput curve is only applicable in continuous processes, like in the series production. To adopt the approach in make to order or engineer to order environments, event discrete effects have to be considering in the computation of the assembly performance.
Johannes Hinckeldeyn; Dennis Kubera; Nils Altfeld; Jochen Kreutzfeldt
Measuring the performance of assembly processes using throughput curves
15
International Association for Management of Technology
IAMOT 2010 Proceedings
Acknowledgements
The authors would like to thank Harburg Freudenberger Maschinenbau, Voith Turbo BHS Getriebe,
General Electric Sensing and Inspection Technologies, PSIPENTA GmbH and Berghof System e.K. for
providing a funding contribution towards the research carried out in the DePlaVis project.
References
Aggarwal, V., Tikekar, V.G. & Hsu, L., 1986. Bottleneck assignment problems under categorization.
Computers & Operations Research, 13(1), 11-26.
Bechte, W., 1984. Steuerung der Durchlaufzeit durch belastungsorientierte Auftragsfreigabe bei Werkstattfertigung,
Boysen, N., Fliedner, M. & Scholl, A., 2007. A classification of assembly line balancing problems. European Journal of Operational Research, 183(2), 674-693.
Chauvet, F., Herrmann, J. & Proth, J., 2003. Optimization of cyclic production systems: a heuristic approach. IEEE Transactions on Robotics and Automation, 19(1), 150-154.
Chen, C. & Chen, C., 2009. A bottleneck-based heuristic for minimizing makespan in a flexible flow
line with unrelated parallel machines. Computers & Operations Research, 36(11), 3073-3081.
Ching, S., Meerkov, S.M. & Zhang, L., 2008. Assembly systems with non-exponential machines:
Throughput and bottlenecks. Nonlinear Analysis: Theory, Methods & Applications, 69(3), 911917.
Cicirello, V.A. & Smith, S.F., 2004. Wasp-like agents for distributed factory coordination. Autonomous
Agents and Multi-agent systems, 8(3), 237-266.
Cicirello, V. & Smith, S., 2001. Insect Societies and Manufacturing.
Dekkers, R., 2006. Engineering management and the Order Entry Point. International Journal of Production Research, 44, 4011-4025.
Eppler, M.J., 2009. Wie Entscheider und Experten reden lernen. Harvard-Business-Manager, 31(4), 5058.
Goldratt, E., 1994. Theory of Constraints 1. Aufl., Gower Publishing.
Goldratt, E. & Cox, J., 2004. The goal: A process of ongoing improvement 3. Aufl., Great Barringtion:
The North River Press.
Hopp, W. & Spearman, M., 2000. Factory Physics 2. Aufl., Singapore: McGraw-Hill Book Co.
Jünemann, R. u. a., 1989. Materialfluß und Logistik, Springer.
Kreutzfeldt, J., 1995. Planen mit Bearbeitungsalternativen in der Teilefertigung. Hannover.
Kreutzfeldt, J., 2007. DEPLAVIS - Durchsatzsteigerung im Anlagenbau und –betrieb durch engpassorientierte Planung und Visualisierung, Hamburg: Hochschule für Angewandte Wissenschaften Hamburg.
Johannes Hinckeldeyn; Dennis Kubera; Nils Altfeld; Jochen Kreutzfeldt
Measuring the performance of assembly processes using throughput curves
16
International Association for Management of Technology
IAMOT 2010 Proceedings
Kung, L. & Chern, C., 2009. Heuristic factory planning algorithm for advanced planning and scheduling. Computers & Operations Research, 36(9), 2513-2530.
Lee, H.L., Padmanabhan, V. & Whang, S., 1997. The bullwhip effect in supply chains. Sloan management review, 38, 93-102.
Lotter, B. & Wiendahl, H., 2006. Montage in der industriellen Produktion: Optimierte Abläufe, rationelle Automatisierung 1. Aufl., Springer, Berlin.
Nyhuis, P., 2008. Beiträge zu einer Theorie der Logistik 1. Aufl., Springer, Berlin.
Nyhuis, P. u. a., 2005. Applying Simulation and Analytical Models for Logistic Performance Prediction.
CIRP Annals - Manufacturing Technology, 54(1), 417-422.
Nyhuis, P. & Wiendahl, H., 2003. Logistische Kennlinien: Grundlagen, Werkzeuge und Anwendungen,
Rajendran, C. & Alicke, K., 2007. Dispatching in flowshops with bottleneck machines. Computers &
Industrial Engineering, 52(1), 89-106.
Rampersad, H.K., 1995. State of the art in robotic assembly. Industrial Robot: An International Journal,
22(2), 10 - 13.
Ribas, I., Leisten, R. & Framiñan, J.M., 2010. Review and classification of hybrid flow shop scheduling
problems from a production system and a solutions procedure perspective. Computers & Operations Research, 37(8), 1439-1454.
Salegna, G.J. & Park, P.S., 1996. Workload smoothing in a bottleneck job shop. International Journal
of Operations & Production Management, 16(1), 91 - 110.
Schultheiss, J. & Kreutzfeldt, J., 2009. Performance improvement in production systems through practice-oriented bottleneck management. In Proceedings of the 4th European Conference on Technology Management. Glasgow, S. 1 - 20.
Song, Y. u. a., 2007. Bottleneck station scheduling in semiconductor assembly and test manufacturing
using ant colony optimization. IEEE Transactions on Automation Science and Engineering, 4(4), 569.
VDI, 1990. 2860: Montage-und Handhabungstechnik; Handhabungsfunktionen, Handhabungseinrichtungen; Begriffe, Definitionen, Symbole. Düsseldorf: VDI-Verlag.
Wiendahl, H. & Hegenscheidt, M., 2002. Bottleneck Analysis of Assembly Lines with Characteristic
Curves. CIRP Annals - Manufacturing Technology, 51(1), 15-19.
Wiendahl, H., 2002. Erfolgsfaktor Logistikqualität 1. Aufl., Berlin Heidelberg New York: Springer Verlag.
Wiendahl, H., 1997. Fertigungsregelung - Logistische Beherrschung von Fertigungsabläufen auf der
Basis des Trichtermodells 1. Aufl., München Wien: Carl Hanser Verlag.
Wiendahl, H. & Nyhuis, P., 1998. Engpaßorientierte Logistikanalyse: Methoden zur kurzfristigen Leistungssteigerung in Produktionsprozessen, München: TCW Transfer-Centrum.
Johannes Hinckeldeyn; Dennis Kubera; Nils Altfeld; Jochen Kreutzfeldt
Measuring the performance of assembly processes using throughput curves
17