Proceedings of TMCE 2008 Symposium, April 21–25, 2008, Izmir, Turkey, edited by I. Horv´ath and Z. Rus´ak c Organizing Committee of TMCE 2008, ISBN 978-90-5155-044-3
PROCESS SIMULATION TO MAKE PERSONALISATION ECONOMICALLY VIABLE Claudia M. Eckert Engineering Design Centre University of Cambridge United Kingdom
[email protected]
David C. Wynn P. John Clarkson Engineering Design Centre University of Cambridge United Kingdom {dcw24, pjc10}@cam.ac.uk
Sandy Black London College of Fashion University of the Arts United Kingdom
[email protected]
ABSTRACT The design of fashion products can be customised to better meet the needs and preferences of individual consumers. Although the technology to achieve this is becoming available, it is still unclear whether mass customisation of fashion products will be economically viable. This paper explores how the economics of different customisation approaches in the textile industry can be assessed by simulation of the design customisation process. These processes are modelled using the Applied Signposting Model. By simulating with different task durations and under different iteration scenarios we show how it is possible to assess the cost of alternative customisation options. This approach to exploring the relationship between product customisation decisions and the performance of the redesign process has potential to support similar decisions in engineering domains.
of fashionable renewal, consumers often feel that the market does not offer products which meet their requirements. The increasing pace of fashion also has significant environmental impact through increased use of energy and other resources for production, transport and disposal.
1. INTRODUCTION
One way to address these issues is to personalise garments to better meet consumer needs and preferences – for instance, by customising both style and fit to individuals’ requirements. This is now being achieved by using mass customisation techniques to create made-to-measure fashion products, in which the consumer can personalise a basic design by selecting styles and/or materials. This has been most successful for high-end products, such as specialised sports shoes (Delamore, et al., 2005) and for products where it is possible to automate the adaptation to individual measurements or material types – such as jeans which have a limited range of styles and fabrics (Crawford, 2005). At the time of writing, the customisation of garment fit is becoming technologically feasible due to the increasing capability and usability of body-scanning technology (e.g., Bougourd, 2006).
Globalisation has brought increasingly cheaper fashion to Western markets. Consequently, consumers now buy more clothes which are worn for shorter time periods. However, despite the accelerating pace
Although the technology for mass customisation of garments is now available, it is still unclear whether this will be economically viable. In particular, once the choice goes beyond adaptation of measurements
KEYWORDS Textile design, mass customisation, product-process trade-offs, applied signposting model (ASM)
785
and selection of materials from a pre-determined set, the required re-design effort could be considerable. The impact of this customisation cost is magnified by the low piece cost of garments compared to the unit cost of most engineered products; the design time is thus a far greater contributor to total cost. Furthermore, anecdotal evidence would indicate that designers of low-volume fashion products often agree commissions for customised products at too low a cost because they underestimate the redesign effort that will be required. This paper will argue that design process simulation models can be deployed to estimate the cost of fashion customisation, thereby empowering these designers to make more informed pricing decisions. The research is part of a larger project entitled Considerate Design, which aims to understand how textile design can better consider the needs of both environment and consumer, while remaining economically viable (Black, et al., 2007). A number of authors describe the application of process modelling methods to support the design process in mechanical engineering or construction. However, such approaches have very rarely been applied to the fashion industry, perhaps since fashion design processes do not exhibit the same scale and complexity. In this paper, process modelling will be applied to these much smaller-scale design processes. We propose that the limited scale and greater repeatability of textile design processes allows a more detailed analysis of design process modelling than is possible in complex engineering processes. Such processes are difficult to observe due to their long duration (many years), access limitations, and the rapid rate of process change compared to the duration of individual projects.
1.1. Relevance to engineering design There has been little cross-fertilisation between design research in the engineering and fashion industries. For instance, fashion and textile design processes have rarely been described in terms of activities and information flows, although this is commonplace in engineering design research (see Eckert, 1997 and Eckert, 2006 for exceptions). There are also many lessons which may be transferred from fashion to engineering, since the degree of globalisation and customisation is greater. For instance, as the trend in engineering moves towards standardisation of a large fraction of product components, others are increasingly customised. Accurately assessing customisation costs will therefore become increasingly
786
important in engineering as well as in textile design. Additionally, due to the small scale and repeatability of textile design projects, these processes allow research questions to be explored which would not be possible in the far more complex engineering design processes.
1.2. Paper overview This paper will begin by discussing some achievements and challenges of mass customisation in the engineering and fashion industries prior to arguing that human designers will remain a fundamental part of this process despite technological and methodological advances. Section 3 introduces the Applied Signposting Model and argues that this approach which was developed for modelling the engineering design process may be applied to support fashion designers in evaluating alternative customisation strategies. Section 4 illustrates this by application to model the design customisation process for repeat-pattern knitwear products. Section 5 highlights the implications of the research for engineering design prior to drawing conclusions in Section 6.
2. MEETING CUSTOMER NEEDS THROUGH MASS CUSTOMISATION This section discusses how customer needs are met by mass customisation of fashion products, by comparison to complex engineered products such as automobiles. It also highlights the role of the designer in fashion customisation and argues that such processes will remain designer-led, despite the introduction of new design and production technology. This highlights the need to support fashion designers in assessing the economic viability of their products and thus motivates the remainder of the paper.
2.1. Customisation of complex engineered products Many complex engineered products are highly customisable. For example, most modern automotive companies offer a range of options for each of their car models. They achieve this through carefullydesigned sets of option packages in which various add-ons may be selected by the customer. This is typically coupled with a carefully designed product platform that allows the company to re-use the design of core components across a range of related products (see, e.g., Simpson, et al., 2001). Generating these option packages adds considerable cost to the design and manufacturing process. They also increase production costs through smaller batch
Claudia M. Eckert, David C. Wynn, P. John Clarkson, Sandy Black
sizes of parts and the increased potential for errors during assembly. Computer programs are used to check the compatibility of selected options and the customer can navigate the design space interactively through an online configurator. For example, in a passenger car flexibility is achieved through a platform with would include a common chassis, engine and power-train for a number of different cars, possibly of different brands. For each car different versions are offered, e.g., a petrol and a diesel model might exist. Customers can select from a range of colours and materials for the interior and exterior finishes. If this does not satisfy the customer’s desire for individuality, specialist companies can customise the car further. In the automotive industry the effort involved generating a viable product platform and option package is warranted by the large production volume. The design cost is only a small part of the cost of an individual product. By contrast, there is no business case for developing such a platform for military helicopters, since they are produced in small numbers and the customisation requirements are very broad. New versions of helicopters are therefore generated when a customer approaches the company, at the cost of considerable redesign effort (Eckert, et al., 2004). The automobile and helicopter examples illustrate two ends of the customisation strategy spectrum: investing up-front to allow the cheaper customisation of individual products through automated configuration and/or automated redesign; or to conduct all redesign at the time of each sale.
2.2. Customisation of fashion products In the fashion industry, mass customisation is now being applied to the creation and fit of made-tomeasure products, where basic design types are offered and the customer may choose style or materials. This has been most successful for high-end products such as specialised sports shoes (Delamore, et al., 2005), trainers1 or tailored suits, and products with a relatively limited range of styles and fabrics, such as jeans (Crawford, 2005) or shirts (Byvoet, 2005). For one example, Brooks Brothers offer a ‘select shirt’ service with a choice of 30 fabrics, 3 fit types, 6 collar and 4 cuff styles2 . The specific design of these products is created by the customer by using a computer tool to choose between pre-determined variations of a basic design. 1 2
http://www.nikeid.nike.com/nikeid/index/ http://www.brooksbrothers.com/selectshirts/
The spectrum of garment customisation Compared to many mass-market products, the production runs of traditional garments are typically small and manufacturers produce a large range of designs at any time. Additionally, while supermarket chains might sell a few thousand products of the same design, only a few dozen instances of a designer garment might be produced. Customisation of fashion products occurs on several levels. A basic level of customisation may start with a colour preference or with the addition of decorative but non-structural features (such as embroidery on jeans). Clothing ranges offer colours and sizes produced in predetermined ratios, with information gathered at the point of sale used to trigger ‘quick response’ deliveries from the manufacturer. At the other end of the customisation spectrum is the handmade bespoke suit or the couture evening gown, where tailored fit, individual fabric choice and dedicated service are priced at a premium resulting in exclusivity. These high-quality luxury items are likely to be kept and maintained for longer than ready-towear items, and are often passed down through generations. Customisation approaches between these two extremes include a wider-than-standard range of sizes and procedures for limited adaptation of standard designs. Customisation of knitwear garments In many respects, knitwear is more complicated than other textiles to produce and customise to different materials, styles and/or sizes. The space of feasible designs is non-uniform in that small changes to the design have potential to cause disproportionate problems. For instance, different yarns have different stretch properties, such that a design which knits well in one yarn might not be possible with a different fibre or even colour. If this occurs the knitting machine would need to be reprogrammed. In another example, to adapt the design to different sizes can be problematic, because the proportion of the garment pieces change and the low resolution of knitted stitches does not allow easy modification of details. While some aspects of sizing could be automated (see Eckert, Stacey, 2003), the problem does not yet have a standard solution and thus requires considerable effort by designers. These examples illustrate that knitwear design displays many of the characteristics more commonly associated with complex engineering domains, where changes to seemingly insignificant details can lead to
PROCESS SIMULATION TO MAKE PERSONALISATION ECONOMICALLY VIABLE
787
major rework (see Eckert, et al., 2004). The knitwear case study therefore allows us to explore some of the strengths and challenges of process modelling to support more general customisation processes. The economics of garment customisation Perhaps the greatest challenge of garment customisation is to find an economically viable way to generate personalised designs. Due to the low cost of individual garments and the low number of garments in a production run, generating an option package is only viable for highly standardised garments, such as suits, shirts or jeans, which are current success stories for mass customisation. Most other designs would need to be re-designed to meet individual needs, as with the helicopter example above. This raises several questions regarding the economics of garment customisation: • What is the customisation cost? • How much customisation should be offered? • Can the garment compete with an off-the-shelf product? • Is it worth investing more to create a more differentiated product? These are the types of questions that we propose may be explored through simulation of the design customisation process.
2.3. The importance of the designer in the fashion customisation process One way to achieve mass customisation is through design automation. While this has been successful in the cases discussed above, the effort is considerable for each application. To date the adaptation rules for particular products were mostly hand-coded, or – as in the case of the car – carefully designed into the product a-priori. A general solution for creating designs automatically in the style of a particular designer and product genre is an active and extremely challenging research field in which no generally applicable method has yet emerged. For example, the work of the famous 20th century American architect Frank Lloyd Wright has been analysed from multiple perspectives over the years; Koile, (2006) examined the stylistic rules of spatial experiences and Koning, and Eizenberg, (1981) developed a shape grammar of this style. The significant theoretical challenges in simply automating the creation of curves with similar stylistic properties are highlighted by Prats, et al. (2006). In summary, the effort in eliciting rulesets from a corpus of prior designs is enormous. In
788
the case of shape grammars it has played a significant role in increasing our understanding of design practice (Stiny, 2006), but has so far not provided commercially viable customisation solutions. Therefore, we argue that the human designer will play an important role in the garment customisation process into the foreseeable future; and supporting the customisation decisions of such designers will remain an important challenge.
3. THE APPLIED SIGNPOSTING MODEL (ASM) In this section, we briefly introduce the design process modelling and simulation approach used by this research prior to discussing its application to support the textile customisation processes. A number of process modelling approaches have been applied to design, particularly in engineering domains (see Browning, Ramasesh, 2007 for a review). Many of these are based on the Design Structure Matrix (DSM). For instance, Austin, et al. (1999) discuss how such a model can be applied to support the planning of construction processes. For another example, Browning, Eppinger (2002) describe a design process simulation model based on the DSM. Other simulation models, such as GERT (Pritsker, 1966) are based on flowchart notations. This research is based upon the Applied Signposting Model (ASM), a simulation model developed for engineering design and development processes (Wynn, et al., 2006). The framework is implemented in the P3 Signposting software tool developed in the Cambridge Engineering Design Centre.3 It includes both task network (flowchart) and DSM-based modelling. The ASM was used for this research as the authors were familiar with the model and since a stable implementation was readily available. However, the research we present here is not tied to the ASM since the subset of features used in this paper exist in many other simulation models. While all tasks in a design process can potentially reveal rework, these unplanned iterations can often be absorbed by compressing subsequent tasks. As a result, major iterations are typically only entered at certain points in the design process. The Applied Signposting Model therefore differentiates between Simple Tasks (yellow boxes) which transform a given set of inputs into a given set of outputs with no possibility of rework, Iteration Tasks (green diamonds), 3
See http://www-edc.eng.cam.ac.uk/p3.
Claudia M. Eckert, David C. Wynn, P. John Clarkson, Sandy Black
which route the process back to a specific point in the process and Compound Tasks (red boxes) which have multiple outputs to advance the workflow along one of several alternative routes. The context of each task is defined in terms of input and output information (called parameters in the ASM terminology) together with a textual indication of the quality, value or state of the information at that point in time. This is illustrated in Figure 1.
Figure 1 An ASM process model illustrating the basic element types
Models can be organised into hierarchical tasks, so that complex processes can be partitioned into a smaller number of high-level activities each with their own input and outputs. These can then be collapsed to provide a high-level overview or expanded to reveal their detail. Models can be constructed from a library of ‘process building blocks’ developed for the domain of interest. These may be repeated in the model to represent unrolled iterations or the repetition of similar activities in different contexts. Constructing such models has been found useful to allow researchers and designers to develop an overview of their processes (Wynn, 2007). Shortcomings can often be identified by inspection of these models, for instance by identifying tasks which contribute little to the design process. Although such insights may seem trivial, they are often difficult to identify in complex processes where designers are
focused on the product and its representations. In such cases few individuals have an overview of how the design is generated and obvious problems can easily be overlooked. Additional insights can be gained by using the model as the basis for process simulation. The ASM provides a discrete-event Monte-Carlo simulation algorithm to enable such analyses. In overview, the algorithm assumes that each task is attempted when its input information becomes available (or is updated following iteration). The task then executes with duration selected from a probability density function. Upon completion, the task’s outputs are updated. This then allows any successor tasks which require that information to be attempted. When an iteration task or compound task is completed, one of the multiple output possibilities is selected according to that task’s configuration. For example, an iteration task governing an iteration cycle may be parameterised with a probability density function indicating the number of iterations expected, or with the likelihood of iteration occurring. The full simulation algorithm (which includes additional features not used in this paper) is described in Wynn (2007). Obtaining accurate estimates for task durations and iterative behaviour increases the burden on knowledge acquisition significantly. Similarly, and ensuring the logical consistency and computability of the model can require significant effort. However, this can be beneficial in a qualitative sense as it forces the modeller to reflect further upon the process and the suitability of the representation. For instance, if it proves difficult to ascertain durations and resource profiles for a task, this would indicate that the way tasks were described did not reflect the way designers view their processes and that the model might therefore be inappropriately structured.
4. PROCESS SIMULATION TO SUPPORT PERSONALISATION This section discusses a model of the process for developing a knitwear collection and its application to support designers in evaluating alternative customisation strategies. The model was constructed by the first author following observations and interviews in 26 companies in the UK, Germany and Italy (Eckert, 1997).
4.1. A generic model of the knitwear design process The original process model comprised fourteen semiformal flowcharts at three levels of detail. The model
PROCESS SIMULATION TO MAKE PERSONALISATION ECONOMICALLY VIABLE
789
has about 140 tasks and 85 decision points. The toplevel process comprises the three stages of Research, Design and Sampling. It is shown in Figure 2.
Figure 2 illustrates that the design process for a collection of knitwear garments is extremely iterative. Designers revisit many tasks to increase their understanding and evaluate decisions they have already taken. For instance, in the Design stage of Figure 2 they concentrate on the generation of ideas for aspects of garments and/or entire garments. Swatches (small fabric prototypes) are produced and evaluated. These swatches may then be modified if designers are not satisfied or if they develop new ideas. These iterations are localised within the Design stage. In the subsequent Sampling stage, entire garments are produced and assessed. The individual designs typically go through several iterations here, but the designers might also question the entire collection. While they would not start the entire collection again from scratch, they might revisit the first stage of the process to conduct more research and/or select alternative materials. For the purposes of this paper, the top-most level of the original flowchart model was simplified to form the basis of an ASM simulation model. This required some modification to ensure the model was logically consistent and appropriately structured for simulation. The resulting model comprises 29 tasks and 9 decision points and is shown in Figure 3.
4.2. Using the generic model to investigate specific knitwear customisation processes
Figure 2 A flowchart of the customised knitwear design process, showing the first of three levels of detail (Eckert, 1997)
This model is generic to a considerable degree of detail (Eckert, 2006). It depicts both the design of an entire knitwear collection and the subsequent design of a particular garment. This is possible because the design process of each individual garment repeats the steps that the entire collection went through, even if only to check whether design assumptions are still valid. For example, designers invest considerable effort to research fashion trends and competitor garments at the beginning of the development of a collection, but systematically revisit this information for the design of each specific garment to ensure that their initial assessments were correct and are still appropriate.
790
The model presented in the previous section is generic in that it represents the task sequence of the knitwear design and production process in most companies. Eckert (2006) argues that differences in the processes of different companies lie in the durations of tasks and the characteristics of design iterations (e.g., the number of iterations undertaken and the time expended in each refinement). The variation is caused by different types of garment, determined by the position the company has in the market and influenced by current fashion trends. While all companies go through the same stages, the more expensive brands typically invest more time into research – spending more effort in the early stages of the process and conducting more iterations to further refine their designs. To use this model as the basis for investigating the customisation of a particular design or collection of designs by a specific company, it therefore needed to be parameterised and adapted. Once parameterised appropriately, the model may be used to assess the
Claudia M. Eckert, David C. Wynn, P. John Clarkson, Sandy Black
which it is better to design each garment individually or to prepare a range of pre-designed sizes, where the closest fit to a customer’s requirements could either be used as-is or adapted slightly. In the following sections, we introduce a specific example of this customisation strategy problem and illustrate how it can be investigated using process simulation.
4.3. Example: repeat-pattern sweaters
Figure 3 An ASM model of the customised knitwear design process. Yellow rectangles represent Simple Tasks; Green diamonds and red rectangles represent Iteration Constructs and Compound tasks respectively. These tasks may fail and drive rework. Blue ellipses represent the information which is generated or required by each task. For full details of the numbered tasks refer to Table 2 of the appendix.
economics of a proposed customisation strategy. For instance, designers can explore the conditions under
In the UK during the 1980s, many independent knitwear designers specialised in made-to-measure, repeat-pattern sweaters of the type shown in Figure 4 (this type of garment was fashionable at the time). They would expect their customers to arrive with particular colours in mind, sets of measurements and possibly with requests for a particular pattern and style – instead of a jumper with a repeating black swan pattern, the customer might request blue cats and extra long sleeves. Traditionally the designer would identify the changes which were required to their original design, set up the knitting machine then knit a new jumper on an industrial knitting machine (or, as still occurred in the 1980s, on a small domestic machine). The main design challenge of such repeating-pattern sweaters was to work out the design such that motifs were not cut in an unsightly way, or that such problems were not visible. If the same design was adapted to different measurements, the designer would need to recalculate the pattern placing to align horizontal and vertical repeats. If this could not be achieved easily, they would consider altering the spacing between the motifs or the length of the garment. If this proved impossible they might try to arrange the pattern such that the motif was cut under the arms or at the neck-line. If the customer had requested a new motif, the designer would first chart the pattern on graph paper then look for another design of similar width and height to form the basis of the redesign process. This customisation process can be fiddly and timeconsuming. It can take several hours, even for this simple example. In the extreme case the designer might make little or no profit from a particular customisation, or might be unable to meet their promised delivery dates. The designer would naturally wonder how much of this customisation process could be shared across similar designs and perhaps conducted up-front.
PROCESS SIMULATION TO MAKE PERSONALISATION ECONOMICALLY VIABLE
791
length measurements (short, medium, tall and extra tall) with four width settings each (slim, medium, large and extra large), such that each design is generated in 12 variants. The setup cost of generating these variants is CV S , where CV S > CBS . The customer has the following options: a) The design is not customised further. The best-fitting off-the-shelf variant in the desired colour is selected and sold to the customer. This incurs no per-garment customisation cost, i.e., CV C = 0. Figure 4 The main challenge of designing a patterned sweater is to ensure that motifs are not cut in an unsightly fashion (Eckert, Stacey, 2003)
4.4. Customisation costs Although this particular design customisation problem could be automated (for a discussion of the challenges see Eckert, Stacey, 2003) it provides a useful example to illustrate the potential of process simulation to support the fashion designer. In this example, it is assumed that the designer expects to sell a number of similar garments which are either customised either a) in color or b) in color and fit. They wish to evaluate the merits of two customisation strategies: 1. A bespoke garment is generated for each customer at the point of purchase, i.e., the entire design adaptation and manufacturing process is conducted. This is based on existing research conducted at setup cost CBS . An additional pergarment cost then depends on the degree of customisation required: a) A standard design knitted in a custom colour scheme. No design effort is required, but the machine must be configured and the garment knitted. The cost of this for one garment is CBC . b) A standard design is customised for the customer’s individual body measurements and knitted in the desired colour. No additional design effort is required, but the create cutting pattern, create fabric sample and pattern placing tasks must be revisited, along with the associated design iterations. The pergarment cost of adapting the measurements is CBM . 2. A standard design is generated up-front in 4
792
b) The best-fitting variant is selected and customised to fit, then knitted in the chosen colour. This is similar to 1b above, but is likely to require fewer design iterations since it is possible to modify a variant which is closer to the final garment. It incurs a pergarment cost CV M where CV M < CBM . The total cost for each of the two customisation strategies may thus be calculated, given the total number of garments N which will be produced and the fraction of customers α who require customised measurements in addition to customised colour: CB(total) = CBS + N [(αCBM + (1 − α)CBC ] (1) CV (total) = CV S + N [αCV M + (1 − α)CV C ]
(2)
Assuming that the customer pay a fixed price for the garment regardless of whether measurement adaptation is required, if the six values of CBS , CBC , CBM ,CV S , CV C and CV M can be identified it is straightforward to identify the best customisation strategy for a given size of production run and fraction of customers requiring measurement customisation. In the following section, we illustrate how these cost components may be estimated and the best strategy identified through simulation of the ASM process model detailed above. Estimating customisation cost using simulation To illustrate this approach, the process model of Figure 3 was parameterised to represent the setup process and the two alternative per-garment customisation processes for each of the two strategies outlined above. The data for this table is based upon the extensive case studies reported by Eckert, 1997), in which a number of knitwear designers and companies were observed. The duration of most design tasks and
Claudia M. Eckert, David C. Wynn, P. John Clarkson, Sandy Black
Variant customisation
49
42
35
28
alpha
350 300 250 200 150 100 50
Inserting the modal values from Table 1 into Equations 1 and 2 above allows calculation of the modal customisation cost for each of the two strategies for any values of N and α. These functions are plotted in Figure 5 (top) and Figure 5 (bottom). These results highlight that, while the bespoke strategy is always the best choice for low production volumes, a point is reached where it is more costeffective to invest in up-front generation of variants; this is represented by the intersection of the two surfaces. The number of garments required to make the latter strategy viable depends upon the value of α, as well as the process-specific cost components. To illustrate, consider Figure 6, which shows crosssections of the two surfaces for α = 0 and α = 1. These charts highlight the modal case only. An optimistic estimate of the smallest N for which the variant strategy would be worthwhile could be generated by repeating the analysis using the maximum values of CBS , CBC and CBM from Table 1, and the minimum CV values. Likewise, a pessimistic estimate could be identified using the maximum CV values and the minimum CB values. The difference between these estimates would highlight the uncer-
42
0 48
Numbe r of garments
36
30
24
0.8 18
CV M 3.0 4.0 7.7
6
CV C 0.0 0.0 0.0
0
0
CV S 69.6 103.2 156.0
12
Min Mode Max
Bespoke design for each customer CBS CBC CBM 12.8 2.5 3.7 28.8 2.5 7.0 76.8 2.5 45.6
0
Number of garme nts
Total cost
Table 1 Summary of the simulation results shown in Figure 5. Values indicate cost in designer-hours
21
Monte-Carlo simulation of the process model for each of the six configurations revealed the profiles of process duration shown in Figure 7 of the appendix. Table 1 summarises these distributions in terms of the minimum, modal and maximum values for each case. These were broadly in accord with the empirical study on which the model was based.
14
0.9 0
Results
400 350 300 250 200 150 100 50 0 7
Total cost
the numbers of iterations required cannot be specified exactly as it depends upon the specific customisation case; therefore such values were specified using probability density functions to indicate the range and distribution of possible outcomes. These six configurations of the simulation model are detailed in Table 2 of the appendix.
alpha
Figure 5 The modal cost of garment customisation for the bespoke strategy (top) and variant modification strategy (bottom)
tainty in the guidance which arises from the uncertain simulation input values.
5. DISCUSSION AND IMPLICATIONS FOR ENGINEERING DESIGN The analysis outlined above is based on simulation of an extremely simple process with few tasks and influencing variables, which is almost trivial when compared to many engineering design processes. Nevertheless, it does illustrate three key points: 1) it is possible to apply process modelling and simulation techniques to the fashion design process; 2) simulation methods have potential to inform decisions about the configuration of such processes; and 3) such methods may be based on a generic design process model which is parameterised to reflect the context of the particular problem. The latter finding highlights the potential for a support system based upon these ideas, since it indicates that the complexity of process modelling and simulation could be hidden behind a user interface allowing designers to
PROCESS SIMULATION TO MAKE PERSONALISATION ECONOMICALLY VIABLE
793
400 350
Total cost
300 250 200 150 100 50 0 0
10
20
30
40
50
Numbe r of garments
Figure 6 Cross-sections through the surfaces in Figure 5 show the intersections between the two surfaces. This plot indicates the minimum number of garments required to justify the variant modification strategy (solid lines) assuming all garments are either measurementadapted (light lines, α = 1) or colour-adapted only (heavy lines, α = 0).
simply enter the estimated durations of tasks in their specific process prior to presenting them with intuitive trade-off charts similar to those above. The approach could also be extended to consider the effect of multiple design processes conducted concurrently; in this scenario, support in assessing the risks of not meeting delivery dates given the designer’s current workload and the variability in individual process’ durations could provide guidance to the designer when deciding whether to take on additional commissions. Such guidance would highlight the non-linear accumulation of schedule risk as workload increases. This is a property of all resource-constrained work queueing systems (see Adler, et al., 1999) and can be difficult to assess without analytical support. Although the example is relatively simple by comparison to many engineering design processes, it is based on a real industrial problem. Fashion processes do display many characteristics of the more complex engineering processes, including complex search spaces, dependence on material variability and the unpredictability of customers, but are not subject to many exogenous uncertainties that cause problems in large engineering processes. Textile processes can thus provide a useful proxy for investigating many issues which are pertinent to understanding and improving processes in engineering domains.
794
In future, we plan to extend this work by conducting case studies to model the design customisation processes for three fashion products which can be personalised using 3D body-scanning: 1) stretchable knitwear garments produced using standard industrial technology; 2) hand-made bespoke bags which are produced in small batches for the high end of the market; and 3) an experimental pseudo-textile project which will create one-off pieces for use at high designer level, with potential for more accessible markets in future. Due to the short duration of these design processes it should be possible to model each process, evaluate the fidelity of the model and assess how it might be adapted to support further design projects. We aim to deploy these models to support the designers in pricing, selecting and conducting their projects. This will allow the feasibility of using process models to support practice to be assessed. This would not be practicable when considering engineering projects of several years’ duration.
6. CONCLUSIONS This paper has shown that it is possible to model and simulate customisation processes in the textile industry by parameterisation of a relatively simple generic model. It is also possible to perform ‘virtual experiments’, although in common with all simulation models their accuracy depends on the quality of the models and the accuracy of time estimates and durations. This analysis shows the potential of using process simulation to explore the impact of customisation decisions which affect the product architecture upon the performance of the customisation process. This is a key issue for customisation processes outside the textile industry. For instance, when designing many product platforms it is important to consider what should be designed up-front and what should be done in response to individual customers’ specific requirements, and how this would be influenced by the expected sales of each variant. In future we will develop more detailed process building blocks for the knitwear case and model other fashion design and production processes. Detailed case studies of specific personalisation processes will be undertaken, in which the durations of tasks performed can be observed. As these processes typically last around one day it is possible to observe many processes in detail and thus develop a rich picture. This will bring the greatest benefit to process modelling in engineering, where observations of en-
Claudia M. Eckert, David C. Wynn, P. John Clarkson, Sandy Black
tire processes are usually not possible due to their long duration and complexity.
REFERENCES
Eckert, C. M., Clarkson, P. J., Zanker, W., 2004, “Change and customisation in complex engineering domains”, Research in Engineering Design, 15(1), pp. 1–21.
Adler, P. S., Mandelbaum, A., Nguyen, V., Schwerer, E., 1995, “From project to process management: An empirically-based framework for analyzing product development time”, Management Science 41(3), pp. 458–484.
Eckert, C. M., Stacey, M. K., 2003, “Knitwear customisation as repeated redesign”, Proceedings of the 2nd Interdisciplinary World Congress on Mass Customization and Personalization, Technical University of Munich, October 2003.
Austin, S., Baldwin, A., Li, B., Waskett, P., 1999, “Analytical Design Planning Technique: A model of the detailed building design process”, Design Studies, 20(3), pp. 279–296.
Eckert, C. M., 2006, “Generic and specific process models: Lessons from modelling the knitwear design process”, Proceedings of TMCE 2006, Ljubljana, Slovenia.
Bougourd, J. P., 2006, “Sizing systems, fit models and target markets” in Sizing in Clothing, Ashdown, S. (ed.), Woodhead Publishing, Cambridge.
Koile, K., 2006, “Formalizing abstract characteristics of style”, AI EDAM, Volume 20, Issue 03, June 2006, pp. 267–285.
Black, S., Eckert, C. M., Delamore, P. 2007, “Developing Considerate Design: meeting individual fashion and clothing needs within a framework of sustainability”, Proceedings of Mass Customization 2007, Boston Browning, T. R., Eppinger, S. D., 2002, “Modeling impacts of process architecture on cost and schedule risk in product development”, IEEE Transactions on Engineering Management, 49(4), pp. 428–442. Browning, T. R., Ramasesh, R. V., 2007, “A survey of activity network-based process models for managing product development projects”, Production and Operations Management, 16(2), pp. 217–240. Byvoet, M., 2005, “Collaborative platform for ecustom fit. Case study shirtdotnet.com”, Proceedings of 3rd MCP World Congress Hong Kong. Crawford, A., 2005, “Leveraging Size UK 3D body data for mass customisation and personalisation. Bodymetrics case study presentation”. Proceedings of 3rd MCP World Congress, Hong Kong. Delamore, P., Junior, V., Lever, G., 2005, “3D direct manufacturing of made-to-measure performance footwear”, Proceedings of Wearable Futures Conference, University of Wales. Eckert, C. M., 1997, “Intelligent support for knitwear design”, Ph.D. thesis, The Open University, U.K.
Koning, H., Eizenberg, J., 1981, “The language of the prairie: Frank Lloyd Wright’s prairie houses”, Environment and Planning B: Planning and Design, 8, pp. 295-323. Prats, M., Earl, C., Garner, S., Jowers, I., 2006, “Shape exploration of designs in a style: Toward generation of product designs”, AI EDAM, Volume 20, Issue 03, June 2006, pp 201–215. Pritsker, A. A. B., 1966, “GERT: Graphical Evaluation and Review Technique”, The RAND Corporation, RM-4973-NASA, April 1966. Simpson, T. W., Maier, J. R., Mistree, F., 2001, “Product platform design: method and application” Research in Engineering Design, 2001. Stiny, G., 2006, “Shape: Talking about seeing and doing”, MIT Press. Wynn, D. C., 2007, “Model-based approaches to support process improvement in complex product development”, Ph.D. thesis, University of Cambridge. Wynn, D. C., Eckert, C. M., Clarkson, P. J., 2006, “Applied Signposting: a modeling framework to support design process improvement”, Proceedings of IDETC/CIE 2006 ASME 2006 International Design Engineering Technical Conferences & Computers and Information in Engineering Conference September 10-13, 2006, Philadelphia, Pennsylvania, USA. DETC2006-99402.
PROCESS SIMULATION TO MAKE PERSONALISATION ECONOMICALLY VIABLE
795
APPENDIX
Figure 7 The customisation costs for the six different scenarios of the knitwear example, as revealed through process simulation. The scales of all histograms’ cost axes are equal and measured in designer-hours.
796
Claudia M. Eckert, David C. Wynn, P. John Clarkson, Sandy Black
Table 2 The parameterisations of the process model shown in Figure 2 which were used to calculate the six components of customisation cost for the knitwear example.
•
In the duration columns, ‘10m 20m 6h’ indicates that the row task’s duration is modelled using a triangular probability density function with minimum duration of 10 minutes, most likely duration of 20 minutes and maximum duration of 6 hours. ‘1h’ indicates that the task’s duration is 1 hour exactly.
•
In the iteration columns, ‘50%, 5’ indicates that there is a 50% probability of rework on any given iteration, with a maximum of 5 attempts. For the tasks which are revisited following iteration refer to Figure 3. Note that the ‘Evaluate knit program & economics’ task #21 is nested within the ‘Generate next variant’ iteration; in this case the maximum number of iterations refers to the maximum per attempt of the outer iteration cycle.
•
All unspecified values are zero. BESPOKE DESIG N FOR EACH CUSTOMER
TASK IN SIMULATION MODEL
CUSTOMISATION OF PRE-DESIG NED VARIANT FOR EACH CUSTOMER
PER-G ARMENT
SETUP
PER-G ARMENT
CUSTOMISATION CUSTOM CUSTOM FIT
CUSTOMISATION CUSTOM CUSTOM FIT
Reesearch
Research
Fashion research in companies
1
Fashion research in retail chains
2
Briefing of designers by buyers
3
Research to understand collection
4
Acquire yarn samples
5
Select yarns
7
Develop design framework
8
Swatch sampling
11 10m20m6h
Evaluate pattern swatches
12 10m20m1d
10m20m1d
13 3h
3h
Swatch sampling for garment selection
14 10m20m6h
10m20m6h
Evaluate garment swatches
15
Detailed design of garments and swatches
Generate technical sketch
16 20m
Evaluate technical sketch
17
10m20m6h
20m
Develop
Create garment shape
18 1m1h6h
1m1h6h
1m1h6h
1m1h6h
knitting
Create garment fabric
19 1m1h6h
1m1h6h
1m1h6h
1m1h6h
machine
P lace patterns
20 1m1h6h
program
Evaluate knit program & economics21
Generate next variant
22
Evaluate garment aesthetics
23
Produce garment
1m1h6h 50%, 5
1m1h6h 50%, 5
1m1h6h 25%, 3
10%, 2
100%,12
Setup machine
24 30m
30m
30m
30m
30m
Knit panels
25 1h
1h
1h
1h
1h
Assemble garment
26 1h
1h
1h
1h
1h
P resent garment to customer
Iterations
Duration
Iterations
9 10
Pattern
Iterations
Duration
Iterations
Duration
Iterations
Duration
10m20m6h
Develop ideas for designs
swatch design
Design
MATERIAL
Swatch sampling for yarn selection 6 10m20m6h
Select design framework
Sampling
Iterations
Duration
MATERIAL
Duration
SETUP
27
Modify collection framework
28
Expectation management
29
PROCESS SIMULATION TO MAKE PERSONALISATION ECONOMICALLY VIABLE
797
798