close

Enter

Log in using OpenID

LAMSADE CAHIER DU 360 - Université Paris Dauphine

embedDownload
lamsade
Laboratoire d’Analyse et Modélisation de Systèmes pour
d’Aide à la Décision
UMR 7243
CAHIER DU LAMSADE
360
décembre 2014
Product variety impact on Bill of Materials Structure
and Master Production Schedule Development:
a case study from the automotive industry
Clément Chatras, Vincent Giard, Mustapha Sali
Product variety impact on Bill of Materials Structure
and Master Production Schedule Development:
a case study from the automotive industry
Clément Chatras*, Vincent Giard*, Mustapha Sali**,
*LAMSADE, **DRM, PSL - Paris-Dauphine University, Place de Lattre de Tassigny F75775 Paris, France
Abstract: The great variety of offerings from companies engaged in mass customization such as in the
automotive sector, implies millions of end products. As in the case of carmakers, this variety proceeds from the
combination of dozens of optional or alternative components whereby thousands or even millions of end products
can be obtained. A number of technical or commercial constraints prevent most combinations of alternative
components (ACs). In these conditions, the creation and updating of the full set of the Bills of Materials
corresponding to each end-product is extremely complex. Additionally, customers may not define their product
requirements by specifying a list of alternative components because of the number of alternative components
(including some entirely unknown to them) and their interdependence. For the purposes of our paper, such variety
is taken for granted and our focus is the associated impact on Bills of Materials. In the past 25 years, scant research
has been dedicated to Bills of Materials representation problems in this context. The most exciting ones use
predicates to formalize combinatorial constraints between ACs and introduce a ‘generic Bill of Materials’ concept
to avoid an exhaustive description of all end products. Additionally, product description for customer purposes was
streamlined through a further concept – the set of alternative services (SAS). The idea of this new approach
developed in this paper, is to describe the product based on its market features, before moving on to the actual AC
combinations based on predicates for alternative services (ASs). We show how some carmakers successfully
use the AS concept to support sales at configurator level and address operational production issues. Because ASs
are also useful for sales forecasting purposes, the concept facilitates Master Production Schedule (MPS)
preparation. Nevertheless, daunting methodological problems remain to be tackled in connection with drawing up
MPSs beyond the frozen horizon: first, owing to forecasting problems in connection with combinations of
interdependent ASs and also because of longer lead times resulting from globalization.
Keywords: mass customization, planning Bill of Materials, generic Bill of Materials, alternative services, master
production schedule
Over five decades, many organizations have switched from mass production to mass customization in order to
better meet customer expectations. This has resulted in an explosion of potential offering variety that far outstrips
actual annual production volume (Pil & Holweg, 2004). This variety of end products which is taken for granted for
purposes of this paper, results from the combination of alternative variants of components. In the case of
sophisticated products, this means that thousands of different components are potentially used on the assembly
line. Standard practice is to group end products comprising a majority of common components into product families,
also called ‘models’. Even when using a modular architecture, offering a huge diversity of sophisticated products
implies managing a large number of alternative components, many of which cannot be combined freely.
Our paper describes the methodological problems arising in the context of huge variety in order to define, store
and retrieve the Bills of Materials required for operational and tactical decision-making purposes. It reviews a
number of solutions that are either used or available to solve these problems. The automotive industry is a perfect
example of this context because a car is the result of the assembly of hundreds of components and few industries
have pushed the envelope further in implementing mass customization (Anderson & Pine, 1997).Therefore, our
paper builds on a real life case to describe both the methodological issues and potential solutions pertaining
situations of high product variety, Accordingly we combine the automotive sector’s technical terminology and that
relevant to the concepts we draw upon in order to fully describe the problems and actual or potential solutions.
For reasons of efficiency, a conventional treatment of this article, starting with a review of literature, was set
aside to better focus on the complex issues addressed. Understanding product variety on an assembly line for a
given product family is a prerequisite to analyzing its consequences for setting up Bills of Materials in the information
systems used for operational and tactical decision making (Section 1). One will then turn to the conventional Bills
of Materials and show (in Section 2) that they are not tailored to an exhaustive description of the population of
millions of different finished products proposed by highly customized mass production industries, by means of a set
of equivalent Bill of Materials references. The arduousness of the task has led researchers to devise solutions that
do away with this need for exhaustiveness. Nevertheless they did not solve all of the commercial and operational
problems one is confronted with. Some carmakers have worked around the problem through a solution hinging on
a functional product description. This enables customers to choose the vehicle they are looking for while enabling
the operational players (in assembly line procurement and production scheduling…), to retrieve the list of ACs
1
required for each vehicle. At tactical decision level, however, the solutions available to describe a vehicle are not
fully satisfactory and researchers don’t seem to have paid much attention to this (Section 3): on the one hand,
building Master Production Schedules (MPS) involves relying on Bills of Materials to procure the required
components for assembly lines, and on the other hand, the sales departments can only base their forecasts on the
functional vehicle features. And translating forecasts based on virtual objects into forecasts applicable to physical
objects to be procured, poses daunting methodological problems. Two valid paths may be explored to solve them,
though none can claim to be a comprehensive solution to the problems.
1. Variety analysis in a mass customization context
From a production point of view, variety is obtained on an assembly line through the combination of optional and
alternative components and a few physical postponement operations (such as painting…). We assume that an
alternative component (AC) is a variant chosen from a set of alternative components (SAC) that may be used in
the same assembly line station. Such variants can differ in terms of technology (engine…), shape (wheels...) or
functionality (radio/radio with CD player…). An optional component (roof bars, parking sensors…) can be said to
be a particular AC of a SAC describing the presence or absence of the component where the implicit and virtual
alternative can be a physical object (bumper without sensor…) or nothing (absence of roof bars…). In the following
pages we only use the general terms of AC and SAC since they cover both the optional and alternative components.
Obviously, end products assembled on a given line share many components but those components that are
systematically included do not impact the end-product variety. Thus, we distinguish two types of components as
shown in Chart 1 representing a sub-set of the SACs. Since the section of the Bill of Materials that describes the
shared components poses no difficulty, we shall only deal with the ACs, which define the identity of the product
chosen by the customer.
Through the commercial information at their disposal, customers cannot be aware of the underlying variety of
the cars they consider purchasing. Indeed, a lot of SACs such as wires or alternators cannot be selected directly
by the customer; such hidden components are indirectly linked to customer choices such as electrical equipment
(air conditioning, heated seats, foldable mirrors, electric sunroof…).
Chart 1: Typology of Components used in Mass Customization
An assembly line variety is sometimes difficult to define because it is closely linked to the definition of a specific
scope including geographic and time dimensions as well as production organization. Those three dimensions
further complicate the definition, updating and use of Bills of Materials. Note that those factors are usually discarded
in papers dealing with variety measurement. So let us address the impact of those three dimensions:
- The commercial offering for a given car model varies from one commercial region to another. A commercial
region (market) can be a country (Germany, UK…) or group of countries (eg:
France/Belgium/Luxembourg/Switzerland). Many products are common to different commercial offerings
but some ACs can be exclusive to a commercial region in order to address special local needs or comply
with local regulations. For example, in the UK the steering wheel is on the right, in Germany pollution
standards are stricter than in France and for the models used in our study, the canvas roof option is only
available in Germany and the smoker pack only available in the UK. Generally, an assembly line produces
vehicles for different markets. Therefore each SAC includes all the ACs of all the markets supplied by the
relevant assembly line. For example, the “roof” SAC of the Renault Twingo must include the “sheet metal
roof”, “sun roof” and “canvas roof” ACs resulting from the consolidation of the German and French varieties.
In practice, therefore, two assembly lines producing the same model for different markets have a different
product variety, but at corporate level, they are merged into a single set.
- Production organization also matters in defining Bills of Material as readily observed when two production
plants assemble the same product family. Indeed, to put in MRP language, a car produced on a line has a
level-0 reference resulting from the assembly of systematic and alternative level-1 components. The ACs of
a SAC are identified by a Bill of Materials reference; that may be a virtual Bill of Materials if the ACs are
produced at an assembly line that is commanded by another that it feeds into, and where no intermediate
stock is required at the station where the ACs are assembled. This convention enables the diversity of a
family produced by multiple lines to exist independently of the industrial organization option (synchronous
production, synchronous procurement). Then there is the case where two assembly lines dedicated to the
same product family adopt different postponement strategies. For example, a plant may assemble an AC
2
made up of an engine coupled to a gearbox while another plant building the same product may receive and
mount the engine and the gearbox separately on two different stations. In this case, one needs to establish
the link between the Bills of Materials and the plants. These organizational aspects are also left aside in the
sections below.
- The features of product offering change over time for three reasons. First, range renewal can involve
changing some ACs. Additionally, AC combinatory limitations may change in line with commercial strategy.
For example, an optional air conditioning system may become standard at a certain point and for a given
equipment level. Finally, for cost or quality reasons some components may be replaced without customers
being aware of this. Therefore the diversity offered by a market is always time-specific and valid only until
further notice. Time-driven change is identified in the literature by the concept of dynamic diversity (Pil &
Holweg, [7]). These temporal aspects are left aside in the sections below.
With these considerations in mind, let us now turn to the impact of production diversity. A variety analysis of a
Renault assembly line shows that vehicle production involves around 700 systematic components and as many
SACs. The 700 SACs account for a total of more than 1900 ACs. Additionally, note that 50% of the SACs are made
up of ASs worth less than €4s. Because of their relatively low cost, one can dispense with MPSs for most of them.
The fact that this number of SACs is greater than the number of stations on the assembly line shows that each
station assembles ACs belonging to multiple SACs. To further illustrate this point, let’s take the example of the
station mounting wheels on the car body: that station will use different nuts depending on the wheel type. Similarly,
for the above reasons, and to further evaluate the volumes involved, note that several dozen Bill of Materials
references for an assembly line dedicated to a model are updated monthly. Finally, two plants that produce the
same model use around 80% of common parts.
Along with the boom of product variety, supply chain globalization driven by global competition has increased
lead times. This dual development has created two new methodological problems that can be addressed by
different solutions. The first is discussed under Section 2 and concerns the definition of Bills of Materials: this has
become difficult due to both the large diversity of end products and the constraints on alternative components
combinations. The second is discussed under Section 3 and flows both from the problem of Bill of Materials
definition and longer lead time in the supply chain. It has to do with the definition of Master Production Schedules
(MPSs) for the final assembly plants and also plants located upstream in the supply chain.
2. Bills of Materials, Mass Customization and Information Systems
All products that can be assembled are described with reference to a Bill of Materials. The Bill of Materials
describes unambiguously a product’s composition and is shared by all departments within the organization
(production, design and manufacturing engineering, sales, management control, after-sales…) (Garwood, 1995).
Each separate product therefore, has a specific reference number in the Bill of Materials. The Bill of Materials used
in production establishes the list and quantity of all components used by an assembly line to obtain a given end
product. For some specific needs, companies may amend Bills of Materials but they are all derived from Bills of
Materials used for production. Both the large variety of products and the combination restrictions between ACs
generate two practical questions: i) how to create the exhaustive list of Bills of Materials for all products potentially
assembled (and does it make sense to do so?)? and ii) how to link a customer’s order (during its definition or during
its production) to a unique Bill of Materials? To answer these questions one needs to analyze Bill of Materials
structure in the information system in the context of mass customization.
We shall first describe the traditional way of building combinatorial diversity into Bills of Materials (section 2.1).
This solution presents problems that some researchers tackled by introducing the concept of generic Bill of
Materials (section 2.2). But due to the current scale of diversity, this solution is no longer adequate. Under section
2.3, we recommend an additional step based on introduction of a different end-product description that reflects
customer added-value. This we call “services” using a functional approach, rather than an organic approach directly
based on components. Because in the production phase, the organic approach is the only valid one, we also
described the solutions required to derive the list of ACs needed for a vehicle from the description of the services
it delivers.
2.1. The Traditional Representation of Combinatory Diversity in Bills of Materials
A solution to the MRP issue of using an AC chosen from a SAC has been available for years. It is based on
planning Bills of Materials, also known as modular Bills of Materials (Stonebraker, 1986). The principle is simple:
the Bill of Materials of an end product is a list of component references systematically included in the product and
a list of dummy component references corresponding to the SACs used on the assembly line. For both real and
dummy components, a Bill of Materials coefficient is used. For example, a car always has four identical wheels: “’4”
is the coefficient for wheels and, because there are several wheels on option, the set of eligible wheels forms a
SAC, represented by a dummy component in the Bill of Materials. In this system, every dummy component must
point to the ACs of the relevant SAC. To precisely define a specific car to be built, in each SAC, a single AC has to
be selected. We shall see below under Section 3 how this structuring is used for planning beyond the frozen horizon.
To isolate the provisional portion of the MPS, one may allocate to the different ACs of a set specific BOM coefficients
standing for the provisional percentage of use of each AC within the relevant set. For the time being, we focus on
3
the fact that a vehicle BOM is defined from the BOMs of its systematic and alternative components, with a single
AC being selected in each SAC, and that all SACs must be taken into account.
In this context, describing an end product in a relational database (Date, 2012) is quite straightforward. The
SACs correspond to entity types, the ACs making up the SACs. Systematic components can be grouped in a kit
representing a specific entity type made up of a single entity. In this approach, the Bill of Materials may be described
through an association of all entity types of the model whose key is a concatenation of the keys of ACs belonging
to different SACs. Note that the combinatory constraints between ACs comprising different SACs introduce
additional complexity in representing diversity.
Those constraints are technical and/or commercial. Technical constraints result either from physical interaction
between components (constraints regarding interface, volume or performance) or from optimization measures in
the purchasing strategy. Commercial constraints result from segmentation to establish consistent offerings and
avoid cannibalization between ranges. Both types of constraints may reflect a pack rationale (compulsory
association of services) or an exclusion rationale. Chart 2 complements Chart 1 by illustrating the respective roles
of both the technical and commercial types of constraints, through a simplified but real case.
Chart 2: Commercial and technical constraints between SACs
In order to describe the problem created by the restrictions, let’s use a virtual car model whose diversity results
from four engines (E1, E2, E3 and E4), two radiators (R1 and R2), three alternators (A1, A2 et A3), three heating
ventilation and air conditioning systems (HVAC) (H1, H2 and H3) and three heater interface panels (I1, I2 and I3).
Excluding the systematic components and the other SACs, free of combinatory restrictions, this diversity of
components yields 4 u 2 u 3 u 3 u 3 216 different end-products. Chart 3 reads from bottom to top. It shows how the
combinatory restrictions between ACs are taken into account step by step. Level 1 associations correspond to
physical constraints, level 2 and 3 associations represent physical and commercial constraints. These restrictions
are drawn from a simplified actual case. We note a legacy barred combinations of level n (mapped with ‘*’) with
level n+1 (mapped with ‘.’). As a result, the actual number of end products drops from 216 to 10 that can be summed
up with the following Bills of Materials: E1R1H1A1I1, E2R1H1A1I1, E3R1H1A2I1, E2R1H2A1I2, E2R2H2A2I2, E3R1H2A2I2,
E3R2H2A3I2, E3R1H3A3I3, E3R2H3A3I3, E4R2H3A3I3.
Chart 3: Example of formal representation using the relational database model
4
Chart 3 illustrates the need to use the 5th normal form of relational databases to describe the existing Bills of
Materials. It highlights the difficulties of taking into account the database integrity issues in connection with the
combinatory restrictions between CAs in the description of a vehicle in the database. When the variety of endproducts is low, as it is for Desktop computers, this approach seems valid.
Where variety results from the combination of dozens of SACs, however, this approach poses both the problem
of defining hundreds of thousands of Bills of Materials describing product variety and the problem of its use in the
definition of the customer’s choices. In practice it is unrealistic to have a customer specify his product by choosing
dozens of ACs (there are too many of them and customers are not even aware of some): another approach is
required.
2.2.
Recourse to Generic Bills of Materials
Hegges and Wortmann (1991) first devised the solution of generic Bills of Materials, and attempted a summary
representation of the set of Bills of Materials comprising an end product family. The generic BOM is based on a
graph combining the systematic components and selected ACs from SACs. The SACs are labeled primary generic
products, and the ACs are selected from parametered values. The solution cuts information redundancy but
requires a prior explanation of the valid combinations between ACs. Van Venn and Wortmann (1992)
complemented this approach by a translation of the constraints into predicates that reduce the number of ACs
combinations. These predicates build on Romanos’ suggested approach (1989). Olsen et al. (1997) then proposed
recourse to programming language to describe the Bill of Materials of model variants thus making Bill of Materials
more legible when restrictions are complex.
Before further analyzing the literature on generic Bills of Materials, let us discuss the use of predicates. A
predicate is a logical expression combining a set of propositions that are either “true” or “false”. The combination
can be a conjunction (AND, noted › ) or a disjunction, (OR, noted š ) and its result is also either “true” or “false”.
This logical sentence serves as a shorthand description of the constraints connecting use of an AC (“true”
proposition) to the presence in the vehicle of ACs belonging to other SACs. Let ACi-1 be the predicate “true” where
the ACi alternative component may be mounted on a vehicle and whose presence depends on that of a set of other
ACj alternative components, chosen from several possible sets. One shall also write ACj (=true) if component ACj
is mounted on the vehicle. One can then write the three following predicates matching the conditions of use of the
three alternators out of fifteen possible parts (each having a predicate) in the example taken in Chart 3:
A11
A21
A21
E1 š R1 š ((H1 š I1) › (H2 š I2 ))
(E3 š R1 š ((H1 š I1) › (H2 š I2 ))) › (E2 š R2 š H2 š I2 )
(E3 š R1 š ((H1 š I1) › (H2 š I2 ))) › (E2 š R2 š H2 š I2 )
In an article that aims to integrate data management of Bills of Materials and routings, Jiao et al. (2000) extended
the Hegges and Wortmann’s (1991) approach of generic Bill of Materials. They proposed a graph made up of nodes
representing physical or abstract entities and lines between nodes to map conjunction constraints (“AND”) and
disjunction constraints (“XOR”). The abstract entity called a ‘generic item’, represents a set of variants (physical
items) of the same type as ACs of a SAC. The graph is also made up of nodes between generic and physical items
in which variety parameters are defined to select “XOR” the appropriate line. Configuration restrictions precluding
free combination of variants are described separately by a set of logical restrictions (predicates) because mapping
them in the graph proved too difficult.
The same year, Bertrand et al. (2000) used a generic Bill of Materials concept called ‘pseudo-item’ to describe
the relevant AC combinations based on given criteria. They proposed structuring the generic Bills of Materials on
the basis of pseudo-items corresponding to a predetermined sequence of customer choices. Each customer choice
is a specific value of the criteria related to the pseudo-items. The outcome is a graph describing the valid
combinations of ACs but it seems only practicable in straightforward cases. Finally they suggested defining the
MPS at pseudo-item level, which, again, is only feasible if there are not too many of them.
Lamothe et al. (2006) have reengineered the generic Bill of Materials concept based on another set of abstract
entities, called logical items. These reduce the variety based on explicit economic considerations. These logical
items serve mainly to describe market segments defined by a set of needs and demand volume. Each need is
translated into a hierarchy of service levels. A service level is matched by a unique indicator for a functional
definition of the product that summarizes a set of criteria and serves to rank them. A market segment requiring at
least one variant is characterized by the list {demand volume, minimum service level}. Each service level can be
satisfied by one or more variants of the end-product. Authors proposed an economic model allowing to determine
production variants meeting all market segment demand at the lowest cost. The cost function includes supply chain
costs whose configuration is driven by the choice of production variants. This model resembles that proposed by
Giard (2000) though the economic functions differ, and does not take explicitly take ACs combination restrictions
into account.
5
2.3.
Bill of Materials Structure Based on Sets of Alternative Services
Lamothe et al. (2006) explicitly introduced the commercial point of view in the definition of valid combinations.
Their structuring approach is similar to that of commercial offering since a customer can view it through a
configurator.
A configurator (Helo et al. 2010, Haug et al. 2012, Trentin et al., 2014) is a web-based interactive application
available to potential customers to make a step by step definition of a customized end-product. The choices made
at different stages can be restricted by previous choices. The configurator reflects the diversity that an average
customer may be aware of. While one could design a configurator enabling customers to customize a product by
choosing all the components, this solution nowadays is impracticable because too tedious and because many
components require in-depth technical knowledge. The solution chosen by most configurators is implicitly based
on the concept of alternative services (ASs). The ASs are grouped into sets (SASs) from which customers can only
choose one AS. The SASs correspond to the logical items used by Lamothe et al. (2006). ASs describe product
functionalities open to customization options in a language customers can readily understand.
The French configurator for Renault Twingo usually works this way, but there is no set sequence such as
‘equipment level’, then ‘motorization’ and ‘transmission’ to end with the various options (e.g., color, alloy wheels,
cooling system and parking sensors). The motorization is a SAS comprising all the main engine features (described
in sales literature for example). It is coupled with a lot of other alternative components (engine, turbo, filters,
sensors, alternator…). The choice of motorization by a customer does not give access to those component variants
as these depend on other parameters. The equipment level indicates the comfort features and motorization services
for a market segment and is based on services and prices. Furthermore, equipment level automatically defines
some components (interior plastics, seat shape and fabric, etc.) whose variety is not displayed. The diversity
featured in the model configurator feeds into a commercial offering comprising eight equipment levels, six engines,
eleven colors, fifteen wheel rims and over twenty-five options. This approach helps closely match customer
expectations in terms of services leaving out actual product architecture and technical definition.
As shown in Chart 4 (drawn from a simplified actual case), SASs are related to technical or commercial
constraints similar to those between SACs in Chart 2. These commercial constraints have the same origin as those
applying to SACs. Technical restrictions between SASs flow from the technical constraints between SACs. This
legacy results from the fact that the choice of AS determines the choice of one or more AC(s) as shown in Chart 5.
Obviously, all these restrictions constraints between SASs can also be described by predicates, as further
discussed below. Despite these restrictions, the diversity of customer offerings remains huge, for example, in
France Renault offers millions of different Twingos.
Chart 4: Representation of set of alternative services (SASs) and their alternative services
This restriction of allowed combinations can be illustrated by a table as table 1 which is based on the real
configurator information.
SAS : Equipement Level (EL)
SAS Motorization (MO) SAS : Transmission (TR)
EL1
MO1
TR1
X
MO2
TR2
.
MO3
TR1
.
MO4
TR1
.
MO5
TR1
.
MO6
TR1
.
TR1 : manual transmission ; TR2 : automatic transmission
EL2
EL3
EL4
EL5
EL6
EL7
EL8
X
.
.
.
X
.
X
.
.
.
X
.
.
X
.
.
X
.
.
.
X
.
.
X
X
X
.
.
.
.
.
.
.
X
.
.
.
.
.
X
.
.
Table 1: Real-life Example of combinatory restriction between ASs of three SASs: Motorization, Transmission and
Equipment Level
6
At this stage, we can draw two conclusions about this configurator:
-
-
The choice of a market (commercial region) is implicit when the customer chooses his country on the web
site before actually customizing the product. The market, therefore can be assimilated to a SAS since it is
chosen by the customer. Note that this choice impacts a number of alternative components that usually
don’t attract customer interest such as the depollution filter (not standard component as it is very expensive).
The car model can also be seen as a SAS because the customer chooses it on the website. Note that
carmakers usually use historical model segmentation to breakdown the large amount of data required to be
stored (for configurator purposes for example). Accordingly models are chosen after the country because
all models are not available everywhere, and customers cannot switch countries without losing all prior
customization settings.
Before describing the impact of AS choice on the determination of the ACs to be used in vehicle production, it is
important to highlight the differences with the example of Chart 1 which illustrates the restrictions applying to
alternative components. Chart 5 shows on a real example that the choice of AC is determined by the choice of one
or more alternative services in as many SASs. We will use the term “determination” to refer to the causal relationship
that links the choice of an AS within an SAS to the choice of one or more ACs. In this approach, it is important to
remember that each SAC is determined by the choice of at least one SAS.
Chart 5: SAC determination by SASs
The determination relation, symbolized by arrows on Chart 5, should be explained. When just a few SASs
determine the choice of an AC, it is possible to use tables to deliver an exhaustive description of the determination
relationship. For example, the three tables below illustrate the combinatory determination of the Alternator (which
is an AC) from two SASs : motorization and cooling system. In the last table, the dashes stand for the exclusions
inherited from the first two tables and the stars stand for additional exclusions flowing from the combination of two
SASs in determining the SAC. In contrast with Chart 3, the choice of alternator is not based on physical SACs
(Engine and Heating system) but rather on virtual commercial SASs: motorization and cooling system.
7
SAS Motorization (MO)
SAC
Alternator
(A)
SAS : Cooling System (CS)
MO1
M02
MO3
MO4
MO5
MO6
A1
X
X
*
*
*
*
A2
*
X
X
X
X
*
A3
*
*
*
X
X
X
SAC :
Alternator
(A)
CS1
CS2
CS3
A1
X
X
*
A2
X
X
X
A3
*
X
X
CS1 : without cooling system ; CL2 : manual cooling
system ; CL3 : automatic cooling system
SAS Motorization (MO)
MO1
SAS Cooling System (CS) CS1
SAC :
Alternator
(A)
MO2
MO3
M04
MO5
M06
CS2
CS3
CS1
CS2
CS3
CS1
CS2
CS3
CS1
CS2
CS3
CS1
CS2
CS3
CS1
CS2
CS3
A1
X
X
.
X
*
.
.
.
.
.
.
.
.
.
.
.
.
.
A2
.
.
.
*
X
*
X
X
*
*
X
X
*
X
*
.
.
.
A3
.
.
.
.
.
.
.
.
.
.
*
*
.
*
X
.
X
X
Tables 2: Example of alternator AC determination based on SAS (Motorization and Cooling system)
The configurator integrates the restrictions between SASs by dynamically restricting the set of possible variants
for the SASs to the subset of variants compatible with previously selected ASs. In fact, a vehicle ordered by a
customer is completely defined by the choice of a single AS from each SAS. The full list of ASs chosen cannot be
seen as a vehicle Bill of Materials because there is normally no one-to-one relation between services and
components. But the list explicitly and easily determines the set of level 1 components to use on the assembly line.
To do so one may adjust the predicates discussed above to determine components.
So let now ACi-1 be the predicate ‘true’ if ACi component is used to assemble a vehicle defined by a set of
alternative services ASi (phrase also used to define the need (ASj=true) for these services in this car). In these
conditions, the three predicates previously used to define the choice of alternators versus other ACs may be
replaced by the following predicates that relate component use to the choice of some alternative services:
A11
A21
A31
(MO1 š (CS1 › CS2 )) › (MO2 š CS1)
(CS1 š MO3 ) › (CS2 š (MO2 › MO3 › MO4 › MO5 )) › (MO4 š CS3 )
(CS2 š MO5 ) › (CS3 š (MO5 š CS6 ))
To assemble a vehicle, the assembly line uses alternator i where ACi-1= true in the relevant SAC. This principle
applied to all SACs serves to evaluate as many predicates as there are ACs in each SAC, where a single AC can
be true because ACs of a given SAC are mutually exclusive. This exploration is not a problem in practice. The
predicates referred to under section 2.2 describe valid combinations of ACs but, unlike those used here, they do
not serve to define the AC to be assembled without referring to the entire Bill of Materials for this particular vehicle.
Note that the alternator is one of those components customers are not aware of, although they are keenly interested
in the motorization and the cooling system option. Thus, the ability to determine the alternator to be used simply by
combining two obvious customer choices is a real benefit of this product description approach combining alternative
services.
This service-based product description involves multiple operational consequences. The first two consequences
address the problems set forth in the introduction of this section 2.
- Based on services, the configurator walks the potential customer through the choice of SACs on offer. In
these circumstances, the guaranty that the customized vehicle will be able to be built (‘ready to buy’) is
obtained by the fact that upstream choices are guided by downstream choices through the dynamic cross
referencing of restrictions between SASs. This approach therefore dispenses with the traditional Bills of
Materials both at configurator and sales department levels.
- Describing a car through a set of built in alternative services serves to establish the ACs it is made of,
through a number of predicates. This solution satisfies short-term practical needs that no longer require
the set of all Bills of Materials one may produce.
- The transcription of the constraints by predicates based on alternative services instead of components
simplifies their definition and updating since there are more commercial constraints than technical ones.
This approach to describing end-products based on SASs and predicates satisfies all practical needs. It seems
difficult to manage the tactical decisions in any other way since, beyond the frozen horizon, production planning is
based on sales forecasts that are defined at SAS level. The problems arise from the fact that MPS have to be
defined at SAC level. This gap is the origin of the major methodological problems that we now turn to.
8
N
3. Production planning in a mass customization context
Drawing up master production schedules (MPSs) for final assembly plants is based on product Bills of Materials.
During the frozen period, the needs of level-1 components are easily deducted from the BOM of the vehicle ordered.
It is interesting to highlight that many carmakers arbitrarily delineate the frozen period without taking into account
the volume of overall backlog that might be important for certain product models. Beyond the frozen horizon, the
definition of the MPS for final assembly plants is more complex. The same is true for MPSs of plants lying upstream
in the supply chain whose orders result from the application of MRP principles based on MPSs of final assembly
pants whose lead times are beyond the frozen horizon.
Therefore, the consistency of production decisions made in the upstream part of the supply chain (final assembly
line) is closely dependent on both the reliability of the MPSs for those lines beyond the frozen horizon and the
stability of this information through the MPS rolling updates. Usually, when the end products variety is relatively
low, MPSs are defined for level-0 items of the BOM, that is to say at end product level. However, beyond a certain
level of variety, it seems easier to define AC MPSs at level-1 of the BOM, as they are fewer. In both cases, the
combinatory constraints between ASs is an issue in defining the MPSs (Section 3.1). If MPSs are defined for level1 components, each AC should be dealt with separately to limit the problem by taking into account only the relevant
constraints (Section 3.2). If the MPSs are defined for end products, then all the constraints must be taken into
account and translated at AC level (Section 3.3).
3.1. Description of the MPS definition problem in the context of mass
customization
In a mass customization context, defining assembly line MPSs requires a prior definition of overall production
volume for each unit of the planning period. Because this type of assembly line is characterized by fixed cycle time,
production volume is directly deducted from overall production time derived from the number of shifts for the period.
In order to manage teams and also part supplies, this information must be forecasted over the planning period, so
we will suppose this to be known for purposes of this paper.
The large number of end products makes difficult to define the MPSs at that level because of the huge
combinatorial between the ACs relevant to them. This difficulty can be lifted as it is not mandatory to use the finest
level of combination for procurement planning. Indeed, the diversity induced by some SACs such as stickers, can
easily be ignored in the planning process provided those ACs are relatively cheap and easy to procure. The
remaining end product diversity is still huge in the car industry, but some automakers use an MPS definition device
at this level; this solution is discussed under Section 3.3. The difficulties of such a device also include some of the
difficulties of level 1 MPS definition for components determined by several SASs.
The definition of systematic level-1 component MPSs is easy as the information (BOM coefficients and total
production volume per period) used for those components is reliable. For the alternative level-1 components, the
planning Bill of Materials presented under Section2.1 and production volume are key to MPS definition beyond the
frozen horizon. Note that a planning BOM is defined for each SAC for a given period: a BOM coefficient is
associated to each AC of the SAC, corresponding to the percentage of occurrence of this AC in the SAC during a
certain period. Assuming that the BOM coefficients are reliably defined – important assumption discussed in the
next paragraph - two MPS calculation approaches beyond the frozen horizon are available; the first one for MPSs
of both end products and level-1 components.
-Note also that the first solution was proposed for the introduction of the planning BOM in the MRP. The
requirements for a period beyond the frozen horizon are also calculated as the total production volume for
the period multiplied by the AC BOM coefficients. This calculation is implicitly based on the assumption of
certain production. With this assumption, calculated demand beyond the frozen horizon cannot be changed
during the rolling planning. This strong assumption is not realistic. In order to limit the cascading spread of
disruptions in the upstream part of the supply chain of the assembly line due to MPS rolling changes, it is
important to dispose of safety stocks for those ACs. Their definition is often tentative, a drawback the second
solution does not have.
-The second solution is based on probabilistic AC demand beyond the frozen horizon. The planning BOM
coefficients for a SAC are used as a vector of a Multinomial distribution in which the number of trials is the
total number of cars produced for the period (Camisulis and Giard, 2010; Sali and Giard, 2012, 2014). AC
demand in a given period beyond the frozen horizon can be seen as a random variable following a binomial
distribution and the safety stock calculation can be done by defining a replenishment level coupled with a
predetermined shortage risk. In this context, it can be shown that the gross requirements for a level-1
component triggered by stochastic MPS requirements beyond the frozen horizon, are random variables
defined as the weighted sum of those random variables (Sali and Giard, 2010, 2014).
The relevance of those two approaches is based on the quality of the planning BOM coefficients’ forecasts. When
the number of SACs is low and there is no constraint between ACs (as it is for desktop computers production), we
can easily define planning BOM coefficients’ forecasts from their demand records. The exponential smoothing
technique can be used when structural change is slow and the lead time relatively short; such forecasts must be
corrected in order to reflect the predictable market trend disruption arising from any new model launches or
9
commercial campaigns by the firm or its competition. If there are constraints between ACs, the use of these
approaches is only possible if the structural characteristics of demand are stable.
The problem is particularly acute in the mass customization context because of product variety and the multiple
combinatory constraints between ACs. In Section 2.3 it was noted that, due to the commercial offering structure
coupled with the impossibility for customers to define demand AC level, it is best to replace the classical BOM listing
ACs for a given vehicle by a list of alternative services. The operational equivalence of those two descriptions is
guaranteed by the determination of each AC using a predicate combining a subset of ASs which describe a product.
This solution fully meets operational production requirements. Let us now turn to a description of AS use to define
MPS for ACs beyond the frozen horizon.
One can easily transpose the SAC planning BOM method for SASs and use the above SAC forecast methods for
SASs. In fact, those methods are more adapted to SASs since they are fewer than SACs and have been defined
to address customer and sales needs whereas SACs correspond to a physical view. Three scenarios illustrated in
Chart 3 should be distinguished:
- If AC procurement lead time is within the frozen period, no MPS needs be drawn up beyond the frozen
horizon. Accordingly, the definition of systematic MPS component is purposeless since it uses known
information ie: production volume and planning BOM coefficient.
- If AC procurement lead time is beyond the frozen horizon, an MPS beyond the frozen horizon is required.
Two scenarios are then possible:
o AC is determined by a single SAS. Here the forecast is straightforward based on the ASs of the SAS.
o AC is determined by multiple SASs. Here, planning BOM definition of AC coefficient is a difficult
problem, even where all AS planning BOM coefficients have been forecasted. Indeed, the existence
in the AC predicate of AS combinations that belong to different SASs is an issue we will address
below.
Number of SASs that determine the SAC
>
”
Frozen Frozen
period period
Lead time
Systematic components
Synchronized supply or
production possible but not
mandatory
Easy supply based only on
the forecasts of production
volume
1
•
Synchronized supply or
production prefered
Synchronized supply or
production prefered
Supply based on the planning
BOM of the unique SAS
Supply based on a complex
replanishement policy that
has to be defined
Table 3: Typology of problems for defining level-1component MPSs
Let us illustrate the forecast definition problems for ACs determined by a predicate using several SASs. Table 4
takes the information of table 2 and serves to highlight how each alternator is defined by a combination of the
motorization services and cooling system services. The three definition predicates for the three ACs of the
“alternator” SAC are shown above the table.
SAS Motorization (MO)
SAS Cooling
System (CS)
MO1
MO2
MO3
MO4
MO5
MO6
CS1
A1
A1
A2
-
-
-
CS2
A1
A2
A2
A2
A2
A3
CS3
-
-
-
A2
A3
A3
A1-1 = (MO1š(CS1 › CS2)) › (MO2šCS1)
A2-1 = (MO2šCS2) › (MO3šCS1›CS2)) › (MO4š(CS2›CS3)) › (MO5šCS2)
A3-1 = (MO5šCS3) › (MO6šCS2›CS3))
Table 4: Determination of alternators via the motorization and cooling system services
If one uses standard probability notation to describe planning BOM coefficients for the ACs and the ASs, one
can draw up table 5 describing alternator calculation coefficients, based on motorization and the cooling system
services; these calculations are shown above the table and use the twelve variables of the table (other than void
and excluding marginal distributions parameters as they are given).
10
SAS Motorization (MO)
MO1
CS1
SAS Cooling
System (CS)
CS2
MO2
MO3
P(MO1šCS1) P(MO2šCS1) P(MO3šCS1)
MO4
MO5
MO6
-
-
-
P(MO1šCS2) P(MO2šCS2) P(MO3šCS2) P(MO4šCS2) P(MO5šCS2) P(MO6šCS2)
CS3
-
-
-
CSSet
P(MO1)
P(MO2)
P(MO3)
P(MO4šCS3) P(MO5šCS3) P(MO6šCS3)
P(MO4)
P(MO5)
P(MO6)
P(A1) = P(MO1šCS1) + P(MO1šCS2) + P(MO2šCS1)
P(A2) = P(MO2šCS2) + P(MO3šCS1) + P(MO3šCS2) + P(MO4šCS2) + P(MO4šCS3) + P(MO5šCS2)
P(A3) = P(MO5šCS3) + P(MO6šCS2) + P(MO6šCS3)
MOSet
P(CS1)
P(CS2)
P(CS3)
100%
Table 5: planning BOM coefficient definition for alternators based on motorization and cooling system services
As 6 P(MOi) and 3 P(CSj) are assumed to be known, defining planning BOM coefficients is actually finding a
solution of the following system with 9 equations and 12 variables.
P(CS1) = P(MO1šCS1) + P(MO1šCS2) + P(MO3šCS1)
P(CS2) = P(MO2šCS2) + P(MO3šCS2) + P(MO4šCS2) + P(MO5šCS2) + P(MO6šCS2)
P(CS3) = P(MO4šCS3) + P(MO5šCS3) + P(MO6šCS3)
P(MO1) = P(MO1šCS1) + P(MO1šCS2)
P(MO2) = P(MO2šCS1) + P(MO2šCS2)
P(MO3) = P(MO3šCS1) + P(MO3šCS2)
P(MO4) = P(MO4šCS2) + P(MO4šCS3)
P(MO5) = P(MOEšCS2) + P(MO5šCS3)
P(MO6) = P(MO6šCS2) + P(MO6šCS3)
We are therefore faced with an underdetermined system that accepts an infinity of solutions, not sufficient to
define the alternator planning BOM coefficients. This under determination is systematic when a SAC is determined
by two or more SASs, except in the rare scenario where each AC of a SAC is determined by a single AS per SAS
that determines this SAC. In order to find a solution to define the alternator MPSs beyond the frozen horizon, we
need to add three additional constraints or add an optimization function. The second possibility is illustrated by
table 6. Both solutions imply a measure of arbitrary decision to solve the problem.
As mentioned above, in a mass customization context, the MPS could be defined for level-1 components
(Section 3.2) or for end products (Section 3.3). We identified two possible uses of planning BOM coefficients to
define an MPS: a determinist approach and a stochastic one. In the MPS definition for level-1 components, we use
a stochastic approach beyond the frozen horizon. In the end product MPS definition, which is the method used by
some carmakers, we use a deterministic approach.
3.2. Defining MPSs for level 1 of the Bill of Materials
Table 3 makes serves to narrow the level-1 MPS component definition problem down by eliminating the
systematic components and those whose lead time does not exceed the frozen period. Thus the problem arises for
the ACs not included in these categories. In this paragraph, it is supposed that these ACs are determined by one
or more SASs.
When one disposes of SAC BOM coefficient forecasts, one may use the probabilistic approach of MPS definition
beyond the frozen horizon (see Section 3.1). To use this solution, one needs SAC BOM estimated coefficients
values. Three cases can be distinguished
-Case 1. SAC BOM coefficients are calculated from the SASs which determine it.
-Case 2. SAC BOM coefficients are calculated without reference to SASs, on the basis of their time series.
In this case, it is useless to make forecasts for SASs that determine these SACs.
-Case 3. The supply of a SAC’s ACs cannot be monitored by the MRP but by a system not based on forecasts
but, for example, on an “order point – fixed ordered quantity” policy. This solution is used to manage the
components whose management is excluded from the MPS, as indicated in the beginning of Section 3.1.
Case 1 – SAC BOM coefficients calculated from SAS coefficients
The first case implies knowledge of BOM coefficients for the SASs that determine the relevant SACs, via the
predicates. Thus, one has the estimated SAS BOM coefficient values required for forecasting SAC BOM coefficients
taken into account in the MPS. One should then distinguish between two possibilities:
- AC is determined by a single SAS. The BOM coefficient forecast for this AC is determined by those of the
concerned SAS. One may then to use the probabilistic approach referred to under Section 3.1, using an
arbitrarily weak risk of stockout, to determine component production launches, required to fully or partly
cover random requirements. This approach was used in about ten benchmarks with real cases drawn from
an automaker (Sali and Giard 2014, Giard and Sali 2012), by comparing the results obtained with those
obtained through the end product MPS solution described under Section 3.3. The random approach yielded
clearly better results.
11
- AC is determined by several SAS. The BOM coefficient forecast problem more complex as showed at the
end of Section 3.1. One can attempt to solve the under-determination problem highlighted at the end of
Section 3.1 through a function to be optimized. Using again the example of table 4, one can attempt to
recourse to estimated values of P0 (CSi š MO j ) to minimize an indicator of difference between these
forecasts and the last values P1(CSi š MO j ) historically recorded. This method is illustrated by tables 6
which are based on the numerical example of table 4, and uses the Min ª¬P1(CSi š MO j ) P0 (CSi š MO j )º¼
criterion to remove the under determination. The left table corresponds to the “historical” table of the joint
coefficients of combined ASs. The right one provides the forecasts conducted on the structures of the
motorization and cooling system services (marginal distributions) and the solution which minimizes the
criterion selected, under constraint of respect of the margins (see equations under table 4). Note that this
method is used here locally and thus differently from what is described under Section 3.3 for the definition
of SAC coefficients, it involves only the relevant SASs and implies that each SAS can be used as often as
necessary. To our knowledge, this approach was never implemented.
2
Initial coefficients P0(CS iš02 M
Forecasted coefficients P1(CS iš02 M
SAS Motorization (MO)
SAS Motorization (MO)
SAS Cooling
System (CS)
MO1
MO2
MO3
CS1
0,020
0,080
0,110
CS2
0,070
0,120
0,060
0,050 0,110 0,080
0,110 0,100 0,090
0,300
0,090
0,200
0,170
0,160 0,210 0,170
1,000
CS3
CSSet
MO4
MO5
MO6
MOSet
MO1
MO2
MO3
MO4
MO5
MO6
MOSet
0,210
0,02667 0,06667 0,10667
0,200
0,490
0,07333 0,10333 0,05333 0,04333 0,09334 0,08333
0,450
0,12667 0,10666 0,11667
0,350
0,100
0,170
0,160
0,170
0,200
0,200
1,000
Forcasted
structure of
cooling
system
Forecasted structure of motorization
P(A1)=0,17 ; P(A2)=0,56 ; P(A3)=0,26
P(A1)=0,1667 ; P(A2)=0,5267 ; P(A3)=0,3067
Tables 6: alternator BOM coefficients example of determination based on motorization and cooling system
services with minimization criterion
Case 2 - Calculation of SAC BOM coefficients based on time series
Let us look at a SAC determined by several SASs. The case of a SAC determined by a single SAS having
implicitly been treated in case 1. The solution used consists in performing direct SAC BOM coefficients forecasts.
Rather than to take the last actual BOM coefficients values for a given SAS, one may use a simple exponential
smoothing technique yielding forecasts Pˆt c!t ( ASi ) D ˜ Pt ( ASi ) (1 D ) ˜ Pˆt 1(ASi ) for these coefficients, for the
periods beyond t. In these calculations, the
¦ i Pˆt c!t ( ASi )
1 constraint must be respected. This is guaranteed at
the initialization of the calculation process of the exponential smoothing which uses coefficients which sum must
be equal to 1; this condition is obtained because this SAC coefficient forecasts use the same smoothing coefficient.
This exponential smoothing technique implies that structure changes are slow and that procurement lead times are
relatively short, failing which, the sales department should conduct an explicit forecast. Two complementary
remarks can be made. The value of the smoothing coefficient used in the SAS BOM coefficients forecasts can be
adjusted periodically, for example based on a minimization criterion of the sum of the squares of the forecast error.
In addition, use of a linear filter always generates oscillations known as the Slutsky-Yule effect, which induce a
skew in the forecasts. This skew can be easily neutralized by increasing forecasts by the quantity corresponding to
the standard deviation of forecast errors. This avoids undervaluation if the forecast is on the high part of an
oscillation (which can be known only latter), at the cost of building excess safety stock in the contrary case.
This approach was subject to a real time three-month benchmark in an assembly plant, on the supply of wiring
harnesses whose SAC mobilizes several SASs; this comparison with the system in place described in the
paragraph below, showed a clear superiority of the probabilistic approach based on direct forecasts of the 17 wiring
harnesses that can be mounted (Camisullis 2008, Camisullis et al. 2010).
3.3. Defining MPSs for level 0 of the Bill of Materials
Some carmakers propose a gradual process that gradually transforms the sales information, expressed at SAS
level, to draw up an MPS at finished product level. To move from SAS level forecasts to finished product forecasts
level, the ASs of the SASs are iteratively combined to gradually expand the definition of an object which, ultimately,
becomes a finished product. The combination of the ASs representing the different SASs gives rise to successive
resolutions of equation systems that are generally under-determined as shown in § 3.1. The final result of this
iterative process, called "enrichment process" by one of the manufacturers that uses this approach, is a forecast
expressed in Completely Defined Vehicles (CDV). Let us briefly examine this iterative process.
At each step of the enrichment process, the resolution of the equation system consists in choosing a particular
solution among the infinite number of possible solutions. To select a particular solution, one needs to define a
selection criterion that may be for example, as proposed in § 3.2, the minimization of a difference between the
forecasts (decision variables) and the last observed values. To pursue the logic of calculating the coefficients at a
higher level of AS combination (eg. combination involving three SASs), MOi š CL j combinations such as
P(MOi š CL j ) ! 0 are considered as ASs that pertain to a set that can be qualified as a Meta-SAS. This iterative
12
process therefore aims to gradually expand, at each period of the planning horizon, the sales forecasts expressed
at the SASs. The tables below illustrate how a calculation of coefficients associated with combinations of ASs from
the SASs Motorization, Cooling system and Commercial Region (prohibited combinations MOi š CL j are not
CR1
Commercial
Region
0,050 0,050
P(CS3‫ٿ‬MO6)
CS‫ٿ‬MOSet
CS3
MO6
P(CS3‫ٿ‬MO5)
P(CS2‫ٿ‬MO1)
P(CS3‫ٿ‬MO4)
P(CS1‫ٿ‬MO3)
P(CS2‫ٿ‬MO6)
P(CS1‫ٿ‬MO2)
CS3
MO5
0,020 0,050 0,070 0,090 0,080 0,060 0,010 0,015 0,060 0,070 0,005 0,070
CR2
CRSet
P(CS2‫ٿ‬MO5)
CS2
MO1
P(CS2‫ٿ‬MO4)
CS1
MO3
P(CS2‫ٿ‬MO3)
CS1
MO2
Initial coefficients
Meta-SAS (Motorization, Cooling System)
CS2
CS2
CS2
CS2
CS2
CS3
MO2
MO3
MO4
MO5
MO6
MO4
P(CS2‫ٿ‬MO2)
CS1
MO1
P(CS1‫ٿ‬MO1)
included in the Meta-SAS combining the SASs Motorization and Cooling system).
0,600
0,040 0,020 0,040 0,080 0,025 0,010 0,080 0,005
0,400
0,020 0,100 0,120 0,090 0,120 0,080 0,050 0,095 0,085 0,080 0,085 0,075
1,000
CR1
Commercial
Region
CR2
CRSet
CS1
CS1
CS2
CS2
CS2
CS2
CS2
CS2
CS3
CS3
CS3
MO1
MO2
MO3
MO1
MO2
MO3
MO4
MO5
MO6
MO4
MO5
MO6
P(CS1‫ٿ‬MO2)
P(CS1‫ٿ‬MO3)
P(CS2‫ٿ‬MO1)
P(CS2‫ٿ‬MO2)
P(CS2‫ٿ‬MO3)
P(CS2‫ٿ‬MO4)
P(CS2‫ٿ‬MO5)
P(CS2‫ٿ‬MO6)
P(CS3‫ٿ‬MO4)
P(CS3‫ٿ‬MO5)
P(CS3‫ٿ‬MO6)
0,027 0,000 0,000 0,073 0,000 0,000 0,000 0,017 0,083 0,127 0,107 0,117
0,067 0,107
CS‫ٿ‬MOSet
CS1
P(CS1‫ٿ‬MO1)
Forecasted coefficients
Meta-SAS (Motorization, Cooling System)
0,550
0,103 0,053 0,043 0,077 0,000 0,000 0,000 0,000
0,450
0,027 0,067 0,107 0,073 0,103 0,053 0,043 0,093 0,083 0,127 0,107 0,117
1,000
Forecasted
structure of
Com m ercial Region
Forecasted structure of (Clim atisation,Motorization)*
*Forecast coefficients obtained at the last iteration (Cf. Tables 6)
Tables 7: Example of BOM coefficient determination for a combination of three ASs belonging respectively to the
Motorization, Cooling System and Commercial Region SASs
The existence of alternative components determined by a large number of SASs involves multiple steps in the
enrichment process. The extreme case is that of an AC determined by all the SASs. In practice, this method leads
to forecasts expressed at finished vehicle level, ie level 0 of the Bill of Materials, excluding any systematic or
postponed components. If it were not improved, the described enrichment process would be difficult to implement
in practice. Indeed, at the last stage of enrichment, the number of decision variables reaches the diversity of finished
products. To work around this problem, ASs are combined into groups of strongly dependent SASs. The notion of
dependency between SASs refers to the existence of technical and/or commercial constraints between ASs
belonging to these sets (cf Chart 4). Meta-SASs formed from interdependent SASs include combinations of ASs
whose coefficients are calculated using the previously illustrated optimization method. By grouping SASs into MetaSASs, which by construction have more limited dependency between them, it becomes possible to calculate
coefficients by combining two by two the elements of the Meta-SASs. The coefficients obtained are then used to
generate, for each period of the planning horizon, CDVs that are characterized over all of the ASs. The total number
of the CDVs per period is equal to the total periodic production of the assembly line. We will return to the device
used to switch from coefficients associated with combinations of ASs to a population of CDVs for each period of
the planning horizon.
In accordance with the time-driven technical and commercial constraints, a CDV is one of the possible AS
combinations representative of all SASs. A CDV has no physical existence, but is a reflection of what could be a
real finished product produced at a given time. It may be considered as a virtual product. By construction, a CDV
respects the commercial and the technical constraints. This property gives the CDV the same consistency as
marketable products. The ACs making up a CDV can thus be easily identified from predicates.
To illustrate how this approach works, take the example of a vehicle that is configurable from the SASs specified
in Chart 5 (Commercial Region, Motorization, Equipment level, Transmission, Driver Side, Wheel rim, Cooling
System, Embedded Computer). To simplify matters, let’s assume a stable commercial offering over the planning
period and a production line dedicated to the assembly of the relevant vehicle model.
In the approach described in this section, the sales department expresses its demand in the form of forecasts for
each AS of all the SASs. These forecasts are expressed as a percentage of total production per period. The sales
forecast enrichment process can then be described in four steps.
13
The first step in the enrichment process is the construction of several Meta-SASs as shown below. It may be
noted that in this arbitrary construction, SASs belonging to a Meta-SAS are interdependent and that the MetaSASs, taken in pairs, are independent except where one of them is that formed from the SASs: Equipment Level,
Commercial Region and Motorization. This particular Meta-SAS, called “principal” and highlighted in blue in Chart
6, constitutes a base from which sales forecasts will be enriched by combining its elements with those of the
secondary Meta-SASs highlighted in yellow in Chart 6.
Chart 6: Meta-SAS definition
The second stage of the enrichment process consists in applying the optimization method referred to above to
each Meta-SAS in order to calculate coefficients for ASs combinations. This yields marginal coefficients related to
new synthetic ASs resulting from the combination of ASs.
The third step also uses the above optimization method (see beginning of the section) to calculate the coefficients
associated with the combination of two ASs, one which belongs to the principal Meta-SAS. For each planning period
the exercise consists in filling Tables 8 below one by one, and in any order where the marginal coefficients are
yielded by the second step of the enrichment process. In Tables 8 the marginal coefficients are materialized by
their respective formula in red.
AS of the principal Meta-SAS (Commerciale Region, Equipement Level and Motorization)
P(CR2 ‫ٿ‬EL6‫ٿ‬MO6)
AS of the secondary Meta-SAS (Embedded
Computer, Cooling System)
MO
P(CR2 ‫ٿ‬EL6‫ٿ‬MO5)
AS of the secondary Meta-SAS (Wheel
rim)
EL
6
P(CR2 ‫ٿ‬EL6‫ٿ‬MO4)
P(CR2 ‫ٿ‬EL6‫ٿ‬MO3)
P(CR2 ‫ٿ‬EL6‫ٿ‬MO2)
P(CR2 ‫ٿ‬EL6‫ٿ‬MO1)
P(CR2 ‫ٿ‬EL5‫ٿ‬MO6)
P(CR2 ‫ٿ‬EL5‫ٿ‬MO5)
P(CR2 ‫ٿ‬EL5‫ٿ‬MO4)
P(CR2 ‫ٿ‬EL5‫ٿ‬MO3)
AS of the
AS of the
secondary Meta- secondary MetaSAS
SAS (Driver
(Transmission)
Side)
CR
6
MO5
2
CR2 EL6
MO4
MO
2
MO
CR2 EL6
6
MO3
EL
EL
1
CR2 EL6
CR
2
CR
6
MO6
2
MO
4
MO
CR2 EL5
5
MO5
EL
EL
P(CR2 ‫ٿ‬EL5‫ٿ‬MO1)
P(CR2 ‫ٿ‬EL5‫ٿ‬MO2)
…
3
P(CRi ‫ٿ‬ELj‫ٿ‬MOk)
CR2 EL5
CR
2
CR
5
…
MO2
2
MO
k
CR2 EL5
j
MO1
…
…
CR2 EL5
EL
…
…
…
P(CR7 ‫ٿ‬EL2‫ٿ‬MO1)
CR
…
MO1
i
CR7 EL2
P(CR6 ‫ٿ‬EL1‫ٿ‬MO6)
P(CR5 ‫ٿ‬EL1‫ٿ‬MO5)
P(CR4 ‫ٿ‬EL1‫ٿ‬MO4)
P(CR3 ‫ٿ‬EL1‫ٿ‬MO3)
MO6
MO
4
MO
CR6 EL1
1
MO5
EL
EL
P(CR2 ‫ٿ‬EL1‫ٿ‬MO2)
P(CR1 ‫ٿ‬EL1‫ٿ‬MO1)
CR5 EL1
CR
4
CR
3
MO2
1
CR2 EL1
MO1
3
CR1 EL1
EC1 CS1 P(EC1 ‫ٿ‬CS1)
EC1 CS2 P(EC1 ‫ٿ‬CS2)
EC1 CS3 P(EC1 ‫ٿ‬CS3)
EC2 CS1 P(EC2 ‫ٿ‬CS1)
Calculation table of the coefficients associated to the combinaisons between an AS
of the principal Meta-SAS and an AS of the secondry Meta-SAS (Embedded
Computer, Cooling System)
EC2 CS2 P(EC2 ‫ٿ‬CS2)
EC2 CS3 P(EC2 ‫ٿ‬CS3)
WR1
P(WR1)
WR2
P(WR2)
WR3
P(WR3)
WR4
P(WR4)
WR5
P(WR5)
DS1
P(DS1)
DS2
P(DS2)
TR1
P(TR1)
TR2
P(TR2)
Calculation table of the coefficients associated to the combinaisons between an AS
of the principal Meta-SAS and an AS of the secondry Meta-SAS (Wheel rim)
Calculation table of the coefficients associated to the combinaisons between an AS
of the principal Meta-SAS and an AS of the secondry Meta-SAS (Driver Side)
Calculation table of the coefficients associated to the combinaisons between an AS
of the principal Meta-SAS and an AS of the secondry Meta-SAS (Transmission)
Tables 8: Calculation tables for the BOM coefficients associated with AS combinations of the principal Meta-SAS
and ASs of the secondary Meta-SASs
14
In the fourth and final step of the enrichment process a set of CDVs that correspond to assembly line total
production for the period is generated, using the coefficients calculated in the third step. To this end, the coefficients
calculated in the third step are transformed into conditional probabilities (probability of having a CDV with an AS of
a secondary Meta-SAS knowing that this CDV is characterized by an AS of the principal Meta-SAS). These
conditional probabilities are used to generate a random integer number of different CDVs. The total number of
generated CDVs must be equal to total assembly line production for the period. The independence of the secondary
Meta-SASs serves to characterize independently a CDV for each one of them. The conditional probabilities are
updated after each draw (draw without replacement) to reflect the changes in the properties of the population. The
products generated and characterized for all SASs correspond to the MPS at level 0 of the BOM. The quantities
generated for the MPS are considered as reliable over the non-frozen period. One is then able to calculate AC
requirements from these quantities through their predicates.
Because they match the physical features of CDVs, MPSs defined at BOM 0 level are comfortable for supply
chain practitioners. Indeed, their use for medium-term planning is treated as a mere extension of the short-term
schedule. By construction, the use of finished product level guarantees consistent component requisitions in terms
of technical and commercial constraints. While finished product level production scheduling is mandatory for the
short-term, conducting this exercise beyond the frozen horizon produces the impression one disposes of reliable
information for the medium-term.
But in practice, the quest for strict consistency through compliance with the technical and commercial constraints,
through finished product level MPSs has substantial consequences on the performance of the upstream supply
chain. The rules inherent in the enrichment process can generate serious disturbances that propagate along the
upstream supply chain, generating bullwhip effect (Childerhouse et al., 2008, Niranjan et al., 2011).
The detailed analysis of the enrichment process used by a French automaker, briefly summarized in this section,
revealed the existence of this effect (Sali 2012). From SAS level planning BOMs provided by the sales department,
the solution selected by the automaker to generate MPSs at finished product level uses mathematical programming
techniques to solve a series of cascading optimization problems for each period of the planning horizon. By
propagating errors in each iteration, this solution ultimately produced unreliable forecasts leading to shortages and
costly emergency supplies, or overstock that impacted cash flow.
4. Conclusion
In the context of mass customization characterized by multiple combinatory restrictions required by this diversity
as regards possible alternative components, Bill of Materials construction and use pose daunting methodological
problems. We argue that detailed Bills of Materials of all potentially manufactured products makes little sense due
to the issue of finding the exact reference of an end product without the list of the relevant alternative components.
Additionally, it is not feasible for customers to fulfill this condition when expressing their needs, due to the technical
knowledge required and the multiplicity of alternative components of which they are unaware. Accordingly,
describing an end-product using a combination of alternative services that correspond to readily grasped functional
features for both customers and sales departments, is the only viable alternative to the traditional Bill of Materials
approach. With this new approach, the list of alternative components for an end product is easily derived from
predicates based on the alternative services selected when defining the end product. This solution meets all
business operational requirements. One must rely on SASs to prepare MPSs beyond the frozen horizon since sales
department forecasts are based on service offering forecasts. Switching from AS- to AC-based forecasts is however
not straightforward when ACs are determined by several SASs. Two MPS formulation options are available and
described in our paper: one, used by several carmakers, boils down to defining MPSs at “level 0” of the Bill of
Materials, the other is based on “level 1”. Both solutions have pros and cons, which calls for further analysis of their
respective performance.
Whatever the preferred solution, both increasing diversity and longer lead times in the supply chain severely
impact forecasting reliability beyond the frozen horizon. This induces unwarranted supply chain costs resulting from
stockout and associated remedial measures. From a strategic standpoint one should focus on the root cause of
these orientations both implemented to increase margin often without proper evaluation of their potential impact on
supply chain management cost. Indeed, increased diversity results from marketing department pressure which
assume that diversity, coupled with pricing differentiation is attractive for customers and profitable for business. On
the other hand, the increase in lead times is due to control department pressure on purchasing costs.
This strategic vision of diversity combined with global alternative components procurement to lower costs has
long been shared by automakers worldwide. But in the aftermath of the 2008 crisis, one notes a paradigm shift in
favor of streamlined offerings based on packages (no automatic cooling system without CD player) or upgrading
strategies (standard fitment of electrical rear window, embedded computer …). In Europe these trends seem driven
by Asian carmakers especially Korean ones. The low-cost rationale has also flourished in recent years and involves
lowering service levels to customers, this strategy also enables the production of less diversified vehicles.
Though all three above trends tend to reduce the diversity of end product offerings, they don’t threaten the mass
customized production model. In fact in the last few years, automakers have been reluctant to cut commercial
offering diversity and have launched different alternative component (AC) streamlining measures while maintaining
the commercial offering. Building modular products is clearly the most successful strategy. By standardizing
15
component interfaces, modularity can substantially lower the number of technical constraints that generate many
of the issues. Also, generalizing assembly line sharing of several models accelerates the delivery of multi-model
platforms that facilitate component sharing. Of all carmakers, the one that seems to have gone furthest in this
standardization rationale is Volkswagen group, which is becoming world no1 carmaker.
To sum up, in order to improve their planning process sustainably in a context of strong diversity, carmakers
dispose of strategic levers on product and upstream supply chain design. Streamlining diversity to limit the cases
of determination involving a multiplicity of steps requires standardization and modularization measures at the
product design stage including ACs. To address technical unfeasibility of perfect modularity involving fully
standardized interfaces, the supply chain design lever may be relied on to switch back to local procurement to cut
lead times beyond the frozen horizon for the ACs determined by more than one SAS.
5. Bibliography
Anderson, D. M., Pine II, J., 1997. Agile Product Development for Mass Customization: How to Develop and Deliver
Products for Mass Customization, Niche Markets, JIT, Build-to-Order and Flexible Manufacturing, McGraw-Hill,
1997.
Aydin, A. O., Güngör, A. 2005. Effective relational database approach to represent bills-of-materials. International
Journal of Production Research 43(6), pp. 1143-1170
Bertrand, J.W.M., Zuijderwijk, M et Hegge, H.M.H., 2000. Using hierarchical pseudo bills of material for customer
order acceptance and optimal material replenishment in assemble to order manufacturing of non-modular products,
International Journal of Production Economics, 66(2), pp 171-184.
Camisullis C. 2008, Les déterminants de la capacité d’une chaîne logistique amont. Thèse de doctorat en sciences
de gestion, soutenue à l’université Paris Dauphine le 3 décembre 2008.
Camisullis, C., Giard, V., Mendy-Bilek, G., 2010, The information to share in upstream supply chains dedicated to
mass production of customized products for allowing a decentralized management, Cahier du LAMSADE, 296.
Childerhouse, P., Disney, S.M. et Towill, D.R., 2008. On the impact of order volatility in the European automotive
sector, International Journal of Production Economics, 114(1), pp. 2-13.
Date, C. J. 2012. Database Design and Relational Theory, O'Reilly.
Garwood D., 1995. Bills of Materials - Structures for Excellence (5e ed,) Dogwood Publishing Company, Marietta
G.A
Giard V. et Camisullis C., 2010. Détermination des stocks de sécurité dans une chaîne logistique-amont dédiée à
une production de masse de produits fortement diversifiés, Journal Européen des Systèmes Automatisés (APIIJESA) 44, pp. 975-1010.
Giard, V. et Sali, M., 2012. Pilotage d’une chaîne logistique par une approche de type MRP dans un environnement
partiellement aléatoire, Journal Européen des Systèmes Automatisés (APII-JESA). 46(1), pp. 73-102.
Haug A., Hvam, L., Mortensen, N.h. 2012. Definition and evaluation of product configurator development strategies,
Computers in Industry, 63(5), pp. 471-481
Helo, P.T., Xu, Q.L., Kyllönen, S.J. Jiao, R.J.2010. Integrated Vehicle Configuration System—Connecting the
domains of mass customization, Computers in Industry, 61(1), pp. 44-52.
Hegge, H.M.H., Wortmann, J.C., 1991. Generic bill-of-material: a new product model. International Journal of
Production Economics, 23(1–3), 117-128.
Jiao, J., Tseng, M.M., Ma, Q., Zou, Y., 2000, Generic Bill-of-Materials-and-Operations for High-Variety Production
Management. Concurrent Engineering: research and applications, 8(4), 297-321.
Lamothe, J., Hadj-Hamou, k., Aldanondo, M., 2006. An optimization model for selecting a product family and
designing its supply chain. European Journal of Operational Research, 169 (3), 1030-1047.
Niranjan, T.T., Wagner, S.M. et Aggarwal, V., 2011. Measuring information distortion in real-world supply chains,
International Journal of Production Research, Vol. 49, n° 11, p.3343 – 3362.
Olsen, K.A., Saetre, P. Thorstenson, A., 1997. A procedure oriented generic Bill of Materials. Computers &
Industrial Engineering, 32(1), 29-45.
Pil, F.K. et Holweg, M., 2004. Linking Product Variety to Order-Fulfillment Strategies, Interfaces, Vol. 35, n° 5, p.
394 - 403.
Romanos M. S.,1989. Demand forecasting for parts used in modular products; a case study, Engineering Costs
and Production Economics, Vol. 17, pp. 231-244.
Sali M., 2012. Exploitation de la demande prévisionnelle pour le pilotage des flux amont d’une chaîne logistique
dédiée à la production de masse de produits fortement diversifiés, thèse de doctorat de l’Université Paris-Dauphine.
16
Sali, M. et Giard, V., 2014. Monitoring the production of a supply chain with a revisited MRP approach, Production
Planning & Control, forthcoming.
Stonebraker, P. W., 1996. Restructuring the bill of material for productivity: A strategic evaluation of product
configuration. International Journal of Production Economics, 45 (1–3), 1 pp. 251-260
Trentin, A., Perin, E., Forza, C.2014. Increasing the consumer-perceived benefits of a mass-customization
experience through sales-configurator capabilities, Computers in Industry, 65(4), pp. 693-705.
Van Venn, E. A., Whortmann, J. C., 1992, New developments in generative BOM processing systems, Production,
Planning & Control, 3(3), 327-335.
17
1/--pages
Report inappropriate content