NEXT GENERATION FACTORY LAYOUTS: RESEARCH CHALLENGES AND RECENT PROGRESS Saifallah Benjafaar Department of Mechanical Engineering, University of Minnesota, Minneapolis, MN 554555 Sunderesh S. Heragu Department of Decision Sciences and Engineering Systems, Rensselaer Polytechnic Institute, Troy, NY 12180 Shahrukh A. Irani Department of Industrial and Systems Engineering, Ohio State University, Columbus, OH 43210 December, 2000 Abstract There is an emerging consensus that existing layout configurations do not meet the needs of the multi-product enterprise and that there is a need for a new generation of factory layouts that are more flexible, modular, and more easily reconfigurable. In this article, we offer a review of state of the art in the area of design of factory layouts for dynamic environments. We report on emerging efforts in both academia and industry in developing alternative layout configurations, new performance metrics, and solution methods for designing the “next generation” of factory layouts. In particular, we focus on describing efforts by the Consortium on Next Generation Factory Layouts (NGFL) to address some of these challenges. The consortium, supported by the National Science Foundation, involves multiple universities and several manufacturing companies. The goal of the consortium is to explore alternative layout configurations and alternative performance metrics for designing flexible and reconfigurable factories. 1. Introduction There is an emerging consensus that existing layout configurations do not meet the needs of the multi-product enterprise [4, 13, 37, 42, 58, 59, 61, 79, 82] and that there is a need for a new generation of factory layouts that are more flexible, modular, and more easily reconfigurable. Flexibility, modularity, and reconfigurability could save factories the need to redesign their layouts each time their production requirements change. Relayout can be highly expensive and disruptive, especially when the entire factory has to be shut down and production stopped. For factories that operate in volatile environments, or produce a high variety of products, shutting down each time demand changes, or a new product is introduced, is simply not an option. In fact, plant managers may prefer to live with the inefficiencies of an existing layout rather than suffer through a costly relayout, which in turn could become quickly obsolete. In our own work with over two dozen companies in the last five years, ranging from big to small, we have encountered mounting frustration with the existing layout choices. This is particularly acute in companies that continuously introduce and offer a wide range of products whose demands are variable and lifecycles short. For these companies, being able to design a layout that can either retain its usefulness over a wide range of product mixes and volumes or be easily reconfigured is extremely valuable. Equally important is designing layouts that can support the need for increased customer responsiveness in the form of shorter lead times, lower inventories, and higher product customization. The current choices of layouts, such as product, process, and cellular layouts do not adequately address the above needs because they tend to be designed for a specific product mix and production volume, both assumed to last for a sufficiently long period (e.g., 3-5 years) [29]. The design criterion routinely used in most layout design procedures - a measure of long-term material handling efficiency, fails to capture the priorities of the flexible factory (e.g., scope is more important than scale, responsiveness is more important than cost, and reconfigurability is more important than efficiency). As a result, layout performance tends to deteriorate significantly with fluctuation in either product volumes, mix, or routings [4, 10, 49, 61, 62, 65]. Using a static measure of material handling efficiency also fails to capture the impact of layout configuration on operational performance, such as work-in-process accumulation, queue times at processing departments, and throughput rates. Consequently, layouts that improve material handling often result in inefficiencies elsewhere in the form of long lead times or large in-process inventories [9]. Hence, there is a need for a new class of layouts that are more flexible and responsive. There is also a need for alternative evaluation criteria for layout design that explicitly account for flexibility and responsiveness. More importantly, there is a need for new design models and solution procedures that account for uncertainty and variability in design parameters such as product mix, production volumes, and 1 product lifecycles. In this paper, we outline the needs and challenges in designing factory layouts in highly volatile environments. We offer a review of state of the art in this area and report on emerging efforts in both academia and industry in developing (1) alternative layout configurations, (2) new performance metrics, and (3) solution methods for designing the “next generation” of factory layouts. In particular, we focus on describing efforts by the Consortium on Next Generation Factory Layouts to address some of the challenges of layout design in dynamic environments. The consortium, founded by the co-authors of this article, is supported by a major grant from the National Science Foundation and involves multiple universities and several manufacturing companies. The goal of the consortium is to explore alternative layout configurations and alternative performance metrics for designing flexible factories. In addition to acquainting readers with results from the initial phase of this effort, we hope to initiate through this article a broader discussion about the physical organization and layout of factories in the future. The paper is organized as follows. In section 2, we review current practice in layout design for factories with multiple products and highlight the limitations of current design methods. In section 3, we review literature on layout design that is pertinent to the central theme of this paper. In section 4, we describe research being carried out under the Consortium for Next Generation Factory Layouts. In particular, we describe results from four streams of research dealing, respectively, with design of (1) distributed layouts, (2) modular layouts, (3) reconfigurable layouts, and (4) agile layouts. In section 5, we report on some emerging trends in industry, in both technology and business practices, that could significantly affect the way factories are organized in the future. In section 6, we offer concluding comments. 2. Current Practice It has been conventionally accepted that, when product variety is high and/or production volumes are small, a functional layout, where all resources of the same type share the same location, offers the greatest flexibility - see Figure 1(a). However, a functional layout is notorious for its material handling inefficiency and scheduling complexity [22, 31, 59, 66, 68, 80]. In turn, this often results in long lead times, poor resource utilization and limited throughput rates. While grouping resources based on their functionality allows for some economies of scale and simplicity in workload allocation, it makes the layout vulnerable to changes in the product mix and/or routings. When they occur, these changes often result in a costly relayout of the plant and/or an expensive redesign of the material handling system [42, 49, 74, 82]. 2 An alternative to the functional organization of job shops is a cellular configuration, where the factory is partitioned into cells, as shown in Figure 1(b), each dedicated to a family of products with similar processing requirements [30, 81]. Although cellular factories can be quite effective in simplifying workflow and reducing material handling, they can be highly inflexible since they are generally designed with a fixed set of part families in mind. The demand levels are assumed to be stable and their life cycles considered sufficiently long. In fact, once a cell is formed, it is usually dedicated to a single part family with limited allowance for intercell flows. While such organization may be adequate when part families are clearly identifiable and demand volumes stable, they become inefficient in the presence of significant fluctuations in the demand of existing products or with the frequent introduction of new ones. A more detailed discussion of the limitations of cellular manufacturing systems can be found in [1, 4, 11, 40, 59, 70, 77]. These limitations resulted in recent calls for alternative cellular structures, such as overlapping cells [1, 39], cells with machine sharing [11, 70], and fractal cells [4, 59, 77]. Although an improvement, these alternatives remain bounded by the underlying cellular structure. (a) Functional layout (b) Cellular layout Figure 1 - Functional versus cellular layout Existing layout design procedures, whether for functional or cellular layouts, have been, for the most part, based on a deterministic paradigm, where design parameters, such as product mix, product demands, and product routings, are assumed to be all known with certainty [29, 54, 56, 72]. The design criterion used in selecting layouts is often a static measure of material handling efficiency (i.e., a total adjacency score or total material handling distance) which does not capture the need for flexibility and reconfigurability in a dynamic environment [9, 10, 13, 43]. In fact, the relationship between layout 3 flexibility and layout performance remains poorly understood and analytical models for its evaluation are still lacking. The structural properties of layouts that make them more or less flexible are also not well understood. Indeed, there exists little consensus as to what makes one layout more flexible than another or as to how layout flexibility should be measured [14, 27, 29, 67, 76, 78, 80]. In turn, this has led to difficulty in devising systematic procedures for the design and implementation of flexible layout. Current design criteria also fail to capture the effect of layout on dynamic performance measures such as congestion, cycle time, and throughput rate. They also ignore the impact of operational parameters such as setup, batching, and loading/unloading at the individual work-centers. More importantly, they measure only average performance and in doing so cannot guarantee effectiveness under all operating scenarios. There are also limitations underlying many of the tools and methods used to design and evaluate factory layouts, making them less effective in factories with high product variety or short lifecycles. We list few of these here based on our own experience with several industry cases [37]. Use of the travel chart as input data: The traditional input data for layout design has been the Travel Chart [73]. However, this chart aggregates the routings and production quantities of all the products produced in a facility. Being a simple graph, it prevents machine duplication analysis. Thereby, it limits the facilities planner to the design of mostly a single type of layout – the functional layout. An alternative could be a Multi-Product Process Chart that captures the unique routing of each product. Such a chart would be essentially a hypergraph representation of the facility that treats each routing as a hyperedge connecting a sequence of departments in the layout. With routing information embedded in the layout, the design of layout configurations, other than the functional layout, becomes possible because partitioning the edge list allows duplication of machines in several locations in the facility [37]. Number of part samples and sampling criteria used to design a layout: A common practice in industry is to use the 80-20 rule (or ABC Analysis) to select one sample of products in designing the entire layout [34]. However, a single sample is rarely an accurate representation for a facility with high product variety or a changing product mix. This problem is compounded by the use of “production volume” as the criterion in selecting a sample. Although a volume-based criterion tends to minimize material handling costs by minimizing material travel distances, it ignores important factors such as revenue generated by each product, frequency of product ordering, and variability in order sizes. The phase in the life cycle of a facility when most models and methods are used: In industry, the dominant use of facility layout design methods tends to be in the midlife or later life of a facility [38]. In other words, facility planners are often engaged in evaluating an existing layout and proposing improvements to it. Clearly, there is considerable opportunity for application of layout algorithms at the conceptual design phase of planning the layout of a facility. Since production data for the entire life of a 4 layout is not known at the initial design stage, there is a need for layout design methods that can work with fuzzy or incomplete data on product mix, routings and production quantities. 3. Literature Review and Classification The facility layout problem has been formally studied as an academic area of research since the early 1950s. Numerous papers on this topic have been surveyed in [6], [48] and [54], among others. In this section, we focus on papers that are pertinent to design of layouts in dynamic environments. We first provide a review of this literature and then offer a possible classification scheme. 3.1 Literature Review Dynamic facility layout: In a 1976 paper, Hicks and Cowan [28] incorporated the costs of relocating departments in analyzing a single period layout. Rosenblatt [63] developed a model and solution procedure for determining an optimal layout for each of several pre-specified future planning periods. This model takes into consideration material handling cost as well as cost of relocating machines from one period to the next. Improvements to the branch and bound procedure in [63] are provided in a number of other papers including [5], [8] and [76]. Heuristic procedure for the dynamic layout problem can be found in a number of papers including [15], [42], [50] and [75], among others. Variations of the basic dynamic layout problem are studied in [7], [57] and [75]. In [57], it is assumed that a goal layout for the last of several pre-specified planning periods can be provided by the designer. A model which uses this goal layout as an input and provides intermediate layouts for the intermediate planning periods is developed. A limitation of this approach is that the relative positions of departments are fixed over all the planning periods - only the sizes and shapes are allowed to vary. For a more complete review of papers on the dynamic layout problem, we refer the reader to [6]. Facility layout in an uncertain production environment: The concept of robustness in analyzing single period layouts was introduced in [65]. Suppose the designer is able to estimate multiple production scenarios for a planning period, for example, optimistic, pessimistic and most likely. A layout is considered to be robust if it performs ‘well’ under all production scenarios. This layout may not be optimal under any specific scenario, but it is also not too far off from the optimal under all possible scenarios [64]. Heuristic strategies for developing robust layouts for multiple planning periods are presented in [47]. Palekar et al. [62] consider uncertainties explicitly in determining plant layout. They formulate a stochastic dynamic layout problem under the assumption that the following are known a priori: (i) material flows between departments for each of several pre-specified planning periods, and (ii) the probability of transitioning from one flow matrix to another. The model is solved via dynamic 5 programming for small sized problems and using heuristics for larger ones. An algorithm for the single and multiple period dynamic layout problem is presented in [46]. Although this method considers additional factors such as additional buffer space and layout changeover costs, it is computationally intractable in the multiple period case. A method for developing “flexible” layouts is presented in [82]. The flexible layout is based on the notion that layouts neither remain static for multiple production planning periods nor do they change during every period. Instead, a layout may remain static for a block of periods, at the end of which the production has changed so much that a new layout is necessary. The question for the designer is not only how to change the layout, but also when to do so. Assuming the flow matrices as well as their probability of occurrence is known for multiple planning periods, the block of periods for which a layout is to remain static is first determined [82]. The layout problem for each block of periods is then solved and results combined to generate a layout plan for multiple production periods. Assuming that future production scenarios along with their probability of occurrence are known, a method for developing multiple period layouts is discussed in [56]. Like the approach in [57], a limitation of this method is that the relative positions of departments are fixed over all the planning periods - only the sizes and shapes are allowed to vary. Distributed Layouts: In order to address the limitations that come from fixed department locations, several authors have recently proposed that functional departments should be duplicated and strategically distributed throughout the plant floor [13, 16, 58]. Duplication would not necessarily mean acquiring additional capacity but could simply be achieved by disaggregating existing departments, which may consist of several identical machines, into smaller ones. Montreuil et al. [58] has suggested a maximally distributed, or holographic, layout where functional departments are fully disaggregated into individual machines which are then placed as far from each other as possible to maximize coverage. Benjaafar [13] has shown that, while some disaggregation and distribution is desirable, full disaggregation and distribution is rarely justified. In fact, the benefits of disaggregation and distribution are of the diminishing type with most of the benefits achieved with having only few duplicates of each department (see section 4.1). Benjaafar also showed that even in the absence of reliable information about product volumes and routings, the simple fact of having duplicates placed throughout the plant can significantly improve layout robustness. Drolet [16] illustrated how distributed layouts can be used to form virtual cells that are temporarily dedicated to a particular job order. Reconfigurable layouts: A shortcoming with several of the above approaches is that they assume production data, including the products to be produced, their routings, type and number of each production resource are known for future planning periods. Even the papers that associate a probability of occurrence with each production scenario implicitly assume that the production resources (type and 6 quantity) remain fixed. In today’s volatile manufacturing environment, it is common to see drastic production changes take place very frequently. It is also common to see old production resources being de-commissioned and new ones being deployed rather regularly. What is challenging for designers is that very often, the changes that are to take place in a production cycle (whether it is change in products, routings, production volume or commissioning and de-commissioning of resources) are known only slightly ahead of the start of the new production cycle. Thus, it seems reasonable for a designer not to look beyond the next period and instead generate layouts that can be reconfigured quickly and without much cost to suit the upcoming period’s production requirements. Heragu and Kochhar [31] discuss this idea and argue that advances taking place in materials and mechanical process engineering, for example, lighter composite materials, nano-technology and laser cutting, will allow companies to reconfigure machines rather easily on a frequent basis. Kochhar and Heragu [42] present a genetic algorithm to solve the associated dynamic layout problem. 3.2. A Classification Scheme In view of the above discussion, we can broadly classify approaches to design of factory layouts for dynamic environments into two major categories. Methods belonging to the first category develop layouts that are robust for multiple production periods or scenarios. Methods belonging to the second develop layouts that are flexible or modular enough so that they can be reconfigured with minimal effort to meet changed production requirements. The first approach assumes that either: (a) the production data for multiple periods is available at the initial design stage itself so that a layout that is robust (and causes minimal materials handling inefficiency overall) over the multiple periods can be identified; or (b) a layout with inherent features (for example, duplication of key resources at strategic locations within the plant) can be developed so that once again, such a layout would help us carry out the material handling functions efficiently through the various production periods. Papers that take the approach outlined in (a) include [47, 61, 62, 65, 69, and 74], among others. A limitation of this approach is that it requires that production data for multiple periods be available at the initial design stage. This requirement is increasingly difficult to fulfill in today’s environment, where factories are plagued by the unavailability of production data for more than one period at a time. Therefore, it is unlikely that this approach - at least on its own - would be adequate to address the needs of factories in the future. The approach described in (b) is more promising since it attempts to build inherent features into the layout that enable it to adapt to changes in the production environment. Papers that take this approach are relatively few and include [13, 16, 42, 58]. 7 The second approach takes the view that layouts would have to be reconfigured after each period. Therefore, the challenge is to design layouts that minimize the reconfiguration cost while guaranteeing reasonable material flow efficiency in each period. Papers that try to balance reconfiguration costs versus material flow efficiency include [5, 8, 15, 41, 49, 57, 63, 74 and 82]. In order to carry out this balancing, this approach requires knowledge of production for each future period. Unfortunately, as previously discussed, this is difficult to satisfy in a volatile environment. A more promising approach is one that attempts to pre-design reconfigurable features into the layout so that reconfiguration costs are always minimal. Very few papers have taken this approach. Some examples include [32], [42] and [43]. Layout design methods for dynamic environment could also be classified based on the design criteria used to evaluate layout alternatives. Much of the literature, including papers that deal with dynamic environments, relies on measures of expected material handling efficiency - a weighted sum of travel distances incurred by the material handling system – in evaluating candidate layouts. Few papers, such as [47] and [65], use a robustness criteria where instead of mean performance, a layout is evaluated by its ability to guarantee a certain level of performance for each period or under each scenario. Others have used a combined mean and variance criterion to minimize the range of fluctuation in performance – see for example [61]. A limited number of papers have considered operational performance as an evaluation criterion. This includes a recent paper by Fu and Kaku [23] who argued that the conventional measure of average travel distances is indeed a good predictor of operational performance, as measured, for example, by expected work-in-process. As we will argue in the next section, this is not always the case. In fact, we will show that in many cases layouts that are designed using operational performance as a criterion can be very different from those that minimize average material handling effort. 4. Next Generation Factory Layouts In this section, we describe research being carried out by the Research Consortium on Next Generation Factory Layouts. The consortium is funded by grants from the National Science Foundation and several industries and involves collaboration between three universities: the University of Minnesota, Ohio State University, and Rensselaer Polytechnic Institute. The goal of the consortium is to explore alternative and novel layout configurations for factories that must deal with high product variety or high volatility in their production requirements. We report on preliminary work undertaken by consortium members in the last two years. In particular, we focus on four promising approaches to layout design that address four distinct needs of the flexible factory. The first three approaches present novel layout configurations, namely distributed, modular and reconfigurable layouts. In the fourth approach, we use operational performance as a design criterion to generate what we term agile layouts. 8 4.1 Distributed layouts The distributed layout concept is based on the notion that disaggregating large functional departments into smaller sub-departments and distributing them throughout the plant floor can be a useful strategy in highly volatile environments. Having duplicates of the same departments, which can be strategically located in different areas of the factory floor, is desirable in a variable environment since it allows a facility to hedge against future fluctuations in job flow patterns and volumes. The distribution of similar departments throughout the factory floor increases the accessibility to these departments from different regions of the layout. In turn, this improves the material travel distances of a larger number of product sequences. As a result efficient flows can be more easily found for a larger set of product volumes and mixes. Examples of departments with varying degrees of department disaggregation and distribution are shown in Figure 2. Such a procedure is especially appealing in environments where the frequency with which product demand fluctuation occurs is too high for a relayout of the plant to be feasible after each change. Thus, a fixed layout that can perform well over the entire set of possible demand scenarios is desirable. Disaggregating functional departments and placing the resulting smaller sub-departments in nonadjoining areas of the layout poses several important design challenges. For example, how should the sub-departments be created? How many should be created? How much capacity should be assigned to each sub-department? Where should each sub-department be placed? How should workload be allocated among similar sub-departments? There are also questions regarding the impact of department disaggregation and distribution on operational performance. For example, how would material handling times, work-in-process, and queueing times be affected? How should material flow be managed, now that there is greater routing flexibility? How should the competing needs for material handling of similar subdepartments be coordinated? There are also important questions regarding what performance measure is appropriate when designing distributed layouts. Should we use a measure of expected material handling cost over the set of possible demand scenarios, or should we use a measure of robustness that guarantees a minimum level of performance under each scenario. More importantly, how sensitive are the final layouts to the adopted performance measure? 4.1.1 Motivating Example Our initial interest in distributed layouts was motivated by work with REI, a leading manufacturer of water filtration products. Their facility was initially organized into 10 functional areas with 4 to 8 workstations per department. The size of the facility and the high diversity of product routings made the distances between individual departments fairly significant. Due to the high product variety and demand volatility, the company found it almost impossible to develop a meaningful layout for its facility. 9 (a) Functional layout (b) Partially distributed layout (c) Maximally distributed layout Figure 2 - Layouts with varying degrees of distribution The need for disaggregating and distributing their large functional departments throughout the factory was initially adopted to reduce the large distances that must be traveled by in-process material. However, it was soon discovered, that when coupled with effective workload allocation, this distribution resulted in significantly lower material handling costs and shorter material handling times even when demand variability was high. 4.1.2 A Layout Design Procedure Some of the above questions are explored in [13, 49]. In particular, we considered situations where the demand for each product is characterized by a finite discrete distribution, represented by a finite number of demand realization scenarios and probabilities of occurrence of each scenario. Demand for each product, characterized by a finite discrete distribution, can be either independent or correlated. The result in both cases is a set of scenarios consisting of different product demand combinations, each with its own probability of occurrence. Characterizing the product demand distributions may be based on historical data and/or forecasts. When the demand distributions are difficult to characterize, equal likelihood can be assigned to the set of possible demand scenarios. Alternatively, the set of scenarios can be aggregated into a smaller subset, which is representative of the range of possible demand realizations. In the case of REI, such an aggregation is possible since orders from different retailers tend to be highly correlated, with order sizes varying over a finite range of discrete order choices. The basic steps of the procedure can be summarized as follows. From (1) the distribution of demand scenarios, (2) the product routings, and (3) the product unit transfer loads, we determine for each possible demand scenario the amount of material flow due to each product between each pair of departments. This results in a multi-product from-to flow matrix for each demand scenario. The objective is to select a layout that provides efficient flow over the entire set of scenarios. For each scenario, we also need to determine the optimal flow allocation among sub-departments of the same type. Thus, we have a combined layout and flow allocation problem. A model for this layout-flow allocation problem, as well as an effective decomposition solution procedure, are given in [13] and [49]. 4.1.3 Some results with a procedure for developing distributed layouts Preliminary experimentation with distributed layouts, using both randomly generated examples and data collected from REI, indicate that significant benefits can be realized by disaggregating and distributing functional departments (over 40% improvement in most cases) [13]. Although the advantage of distributed layouts is most pronounced when demand variability is high, it is significant even in the absence of variability. This is particularly the case for layouts with large departments or a large number of departments. If the distribution of flow patterns can be categorized a-priori, then including flow 10 information at the design stage can lead to higher quality layouts. However, material handling costs can be significantly reduced even if no flow information is included (e.g., by a random distribution of subdepartments). In addition the quality of distributed layouts is quite insensitive to inaccuracies in the demand distribution. But most importantly, most of the benefits of department duplication are realized with relatively few replicates. This means that there would be rarely a need to fully disaggregate functional departments. Having a layout where department replicates are distributed throughout the plant floor can also be effective in handling products with short runs or products with short lifecycles. This can be achieved, for example, by the formation of temporary cells dedicated to a particular product line or customer job order. These cells can be quickly formed, as shown in Figure 3, by finding adjoining replicates that can be temporarily dedicated to a product line or a customer job. This cell is disbanded once the product is phased out or once the customer order is completed. The individual replicates are then free to participate in new cells. An early vision of these virtual cells is also discussed in Drolet [16]. Furthermore, we have found that distributed configurations can be useful in handling growth or contraction in a graceful manner [49]. For example, in many industries product maturity occurs over several periods. Instead of redesigning the facility for each phase of product growth, we found that a distributed layout can significantly minimize rearrangement costs which would be necessary if a functional configuration is adopted. Additional machines are added to the periphery of the existing layout as needed and without necessarily relocating equipment. Growth occurs almost in a concentric fashion that keeps layout space compact and maintains efficient material handling. More importantly, this approach allows adding or subtracting capacity in smaller increments than would be possible otherwise since introducing or removing capacity always takes place at the periphery while maintaining the factory core intact. Virtual cells Figure 3 – Using distributed layouts to construct Virtual Cells 11 4.1.3 Research Challenges Several research challenges remain to be addressed. In our initial effort, we assumed that the number of department copies and the capacity of each copy are known. In practice, these are decisions that facility designers must make before the layout process can be carried out. Our initial models do not account for the cost of disaggregating and distributing departments nor do they capture the economies of scale associated with operating consolidated departments. The infrastructure that is typically shared by a single consolidated department in job shops, such as operators, computer control systems, loading/unloading areas, and waste disposal facilities, must be duplicated in a distributed layout across all department copies. Thus, while there may be material handling benefits to department disaggregation and distribution, these benefits should be carefully traded-off against the advantages of operating consolidated facilities. Therefore, there is a need for an integrated model that combines department duplication and capacity assignment with layout design and flow allocation. In our initial flow allocation model, we assumed full flexibility in assigning workload among duplicates of the same department. In practice, this could result in splitting orders that belong to the same product among several duplicates. This would mean smaller batches and possibly longer and more frequent setups. Order splitting could also cause delays in shipping completed orders due to poor synchronization among individual batches of the same order. Addressing this problem would require either capturing setup minimization in the objective function or placing additional constraints on flow allocation to prevent order splitting. 4.2 Modular Layouts The focus of this approach is on design of customized layouts for facilities with multiple products. We are considering a novel approach based on the idea that layouts can be constructed as a network of basic modules. Here, we assume that, at least in the short term, the product mix is known and demand is relatively stable. As the product mix evolves and demand changes, certain layout modules will be eliminated and others added. The use of modules is motivated by the fact that none of the prevailing layout configurations (functional, flow line, and cellular) can individually describe the complex material flow network in a multi-product manufacturing facility. Preliminary research on this topic was undertaken and has recently been reported in [35, 37-39]. The research sought to answer the following fundamentally new questions: Could an alternative layout other than the three traditional layouts be a better fit for the material flow network in a multi-product manufacturing facility? And, could this alternative layout be a combination of the three traditional layouts? The proposed concept of designing any facility layout as a network of layout modules provides a meta-structure for the design of multi-product manufacturing facilities. The proposed concept uses the idea of grouping and arranging the machines required for subsets 12 of operations in different routings into a specific (traditional) layout configuration that minimizes total flow distances or costs. 4.2.1 Motivation Figure 4 shows an example of a new layout configuration that is being proposed for multi-product manufacturing facilities. It was designed during a study that was done for Motorola Inc. The company wanted to assess the feasibility of changing the functional layout in one of their semiconductor fabs into a cellular layout. The Functional layout in the fab is comprised of seven bays (or process departments) – DIFF, ETCH, FILM, IMPLANT, PHOTO, METROLOGY and BACKEND. Four product routings that were representative of the product flows in the fab were provided for the study. The study found that a cellular layout was not a viable option for the fab. This was because the creation of flowline cells based on grouping one or more routings required significant equipment and process duplication among the cells. However, a visual string matching analysis of the routings revealed that, despite being dissimilar, different pairs of routings had one or more common substrings of operations that were either identical or, at least, had high commonality of operations. Based on this observation, the novel layout shown in Figure 4 was generated. Unlike the functional, flowline or cellular layouts, this layout uses a combination of the three traditional layouts to arrange the equipment in different areas of the facility. In addition, this layout has allowed some machine duplication, as is usually done to design a cellular layout for a multiproduct manufacturing facility. In the layout of Figure 4, all pairs of consecutive operations in all the product routings are performed in (a) the same layout module or (b) adjacent layout modules. A layout module is a group of machines in a portion of the overall facility that has a flow pattern characteristic of a traditional layout. Since this work for Motorola Inc., a study of samples of product routings obtained from published data and from industry was conducted. A common observation was made that dissimilar product routings often had common substrings of operations that could be aggregated into layout modules. 4.2.2 Classification of Layout Modules It appears that the material flow network in any multi-product facility can be decomposed into a network of layout modules, as shown in Figure 4, with each module representing a portion of the entire facility [35, 37]. Each layout module is a group of machines connected by a material flow network with a well-understood flow pattern and method for design of its layout. The initial set of modules we are proposing consists of the three traditional layouts shown in Figure 5. For example, the Flowline and Cell Modules have a part family focus. The Flowline Module is an aggregation of one or more routings that are identical. Whereas, the Cell Module is an aggregation of a family of similar routings based on a 13 Functional Layout for ETCH Flowline for ETCH 2.06 2.08 2.09 2.10 2.07 Flowline for PHOTO 2.01 7.03 5.01 5.02 5.03 5.04 5.05 5.06 7.02 7.01 3.08 Functional Layout for FILM Department 5.07 7.04 7.05 3.04 3.07 2.05 6.01 3.01 3.02 3.04 3.05 3.06 5.02 5.03 5.04 Flowline for BACKEND 5.05 2.02 4.01 1.03 1.04 1.02 1.01 1.05 Cell for ETCH, IMPLANT and PHOTO Functional Layout for ETCH, FILM and PHOTO Flowlines for DIFF Figure 4 - Example of a facility layout designed using layout modules A B C D E A C D E F B G H (a) Flowline Module (b) Branched Flowline Module C B D A E A+B+C (d) Machining Center Module (c) Cell Module B C A D B D C E A E (f) Patterned Flow Module (e) Functional Layout Module Figure 5 - Example layout modules pre-defined level of commonality of machines used and similarity of sequence in which the machines are used. In contrast, the Functional Layout Module is a group of machines that does not process a family of routings. However, the material flow pattern in its From-To chart could correspond to an acyclic digraph, as in an assembly or disassembly line or, in the worst case, a completely connected digraph. 4.2.4 Solution Approach The ideal solution would have each product completely processed on a dedicated flowline. Since that would entail significant investment in equipment, a practical approach would be to maximize the number of consecutive operations in a family of routings that are performed in the same module. In order to realize such a structure, we developed a solution approach based on the methods of string matching and clustering used extensively in genetics, molecular chemistry and the biological sciences [35]. At the core of the approach is the concept of a ‘common substring’ and a ‘residual substring’ in a product routing, defined as follows. Common substring is a substring of consecutive operations that is common to two or more operation sequences; Residual substring is the remaining substring(s) of operations in an operation sequence after all common substring(s) are extracted from it. For example, given two operation sequences Sa (1→2→3→4→7→8) and Sb (1→2→5→6→7→8), the common substrings are 1→2 and 7→8. The residual substrings are 3→4 and 5→6 in sequences Sa and Sb, respectively. In the current version of the approach [37, 39], given the sample of routings for products produced in the facility, the common substrings between all pairs of routings are first extracted. Next, the frequency of global occurrence of each common substring in the routings of all products produced in the facility is computed. This is followed by an aggregation step where similar substrings are aggregated and each cluster of substrings becomes a layout module. This is followed by a disaggregation step where certain modules are eliminated because they do not fulfill a minimum machine utilization criterion or constraints on machine allocation and duplication among multiple modules. Figure 5 shows the typical result expected from this approach – a facility layout that is a network of dissimilar modules, in this case, a Cell Module (M2), two Patterned Flow Modules (M1, M4) a Flowline Module (M3) and a Functional Layout Module (Machine #2) [37]. 4.2.4 Research Challenges Several important research problems need to be solved: (1) Having identified all common substrings, it will be required to aggregate several of those substrings into a single module to minimize machine duplication costs. A measure of dissimilarity and a threshold value for aggregating similar substrings need to be developed. This is related to the problem of determining the optimal number of modules in the final layout. One idea is to develop measures of connectivity and transitivity of the directed graph 14 obtained by aggregating a set of common substrings. (2) Feasibility criteria need to be established for allocating machines to several modules subject to machine availability and minimum machine utilization criteria. An iterative loop needs to be incorporated in the design approach to absorb any module that is rejected based on either or both of these criteria. (3) The current approach treats each residual substring as a sequence of operations performed on machines located in process departments. It seems logical to cluster these substrings and aggregate their machines into cell modules based on user-defined thresholds for string clustering. (4) The performance of this new layout will have to be compared with those of Flowline, Cellular and Functional layouts generated for the same facility. An important part of this activity will be the computation of all costs associated with a facility layout, such as WIP, material handling, queuing delays, setups, and processing efficiency. M1 9 M4 9 7 10 7 8 6 1 4 6 2 5 10 M2 11 7 1 3 M3 12 Inter-module flow or flow between a module and an individual machine Intra-module flow Figure 6 - Facility Layout as a Network of Layout Modules 4.3 Reconfigurable Layouts Our third focus is on design of reconfigurable layouts. Here, we consider the case where resources can be easily moved around so that frequent relocation of departments is feasible. This is motivated by the fact that in many industries (e.g., consumer electronics, home appliances, garment manufacturing, etc), fabrication and assembly workstations are light and can be easily relocated [20, 53]. In fact, even in the metal cutting industry, recent advances in materials and processing technology are making it easier for manufacturing facilities to be configured and reconfigured on a more frequent basis. For example, many discrete manufactured components are made of composite materials that are light in weight and have 15 much better mechanical properties (e.g., vibration absorption properties). Aluminum composites, for instance, can now replace cast iron parts [24] and, phenolics are replacing aluminum parts, among others [2]. Newer processing technologies such as electron beam hardening, molecular nano-technology and laser cutting is resulting in lighter weight machining equipment [3]. Permanent magnetic chucks that carry their own energy source, that do not obstruct machining, and do not magnetize the cutting tool are also being developed [17]. With these developments in materials and processing technology, we are moving towards processing technologies which employ light weight machine tools and can process light weight parts. It is possible to envision facilities where these light weight equipment are mounted on wheels and are easily moved along suitably designed tracks embedded in the shop-floor [28 and 42]. As a result, it may not be too far fetched to say that the layout will be changed several times a year. In fact, through a workshop and a delphi survey, the committee on Visionary Manufacturing Challenges for 2020 [60] has identified adaptable processes and equipment and reconfiguration of manufacturing operations as two key enabling technologies that will help companies overcome two of the six grand challenges or fundamental goals to remain productive and profitable in the year 2020. These grand challenges are to “achieve concurrency in all operations” and to “reconfigure manufacturing enterprises rapidly in response to changing needs and opportunities”. When frequent relayout is feasible, the layout design problem can be significantly simplified even when product demand and product mix are highly variable. It becomes possible to focus only on the immediate product mix and the immediate production volumes. However, since we would typically incur (1) some loss in production capacity during the relocation process, and (2) a relocation cost associated with the physical movement of resources (e.g., labor cost, dismantling and reconstruction costs, rewiring costs, and startup/setup costs), we must account for these costs when deciding whether it is beneficial to remove a resource or leave in its current location. A general design and planning framework for carrying out this process is shown in Figure 7. A model and solution approach for this problem is provided in [42]. The objective function of the model consists of two terms: a material handling cost term and a relocation cost term. The magnitude of the relocation costs determine whether a relayout is carried out or not. In the extreme, where relayout costs are insignificant, an entirely new layout can be generated during each period. On the other hand, if relayout costs are prohibitive, the existing layout would be retained. In practice, the two extreme scenarios would be unlikely. Instead it would be desirable to relocate some of the resources during each period. The layout would then evolve gradually over time as flow patterns evolve. The cost of relayout could be reduced if investments in infrastructure that facilitates relayout are made during the initial design of the factory. For example, the facility may be designed so that it has embedded tracks that help decrease the cost of moving equipment. It may also be possible to design all interface devices for control systems so that they are interchangeable and open. In such a case, “plug and 16 play” features may be implemented at the workstation level. Support services such as compressed gas, water or coolant lines, and waste disposal may have to be suitably designed for the concept. A primary advantage of reconfiguring a layout when warranted by changes in product mix and volume is that material handling cost can be minimized because equipment can be reconfigured to suit the new production mix and volume. Of course, this cost must more than offset the cost of moving equipment from its current location to a new one. In addition, due to the short term life of a given layout and production data availability for this time period, it is possible to consider optimizing operational performance measures such as minimizing part cycle times, work in process inventory, or throughput. The potential to frequently alter layouts, therefore, transforms the modern layout problem from a strategic problem in which only long term material handling costs are considered to a tactical problem in which operational performance measures such as reduction of product flow times, work in process inventories, and maximizing throughput rate are considered in addition to material handling and machine relocation costs when changing from one layout configuration to the next. A framework for the reconfigurable layout problem as well as methods for estimating performance measures of such a layout are provided in [29] and [30], respectively. A method for designing layouts with operational performance in mind is given in the next section. Design Data + New Product Design + New Processes Selected Production Data + Expected Volume + Changed Product Mix Revised Material Flow Matrices / Adjacency Matrices Material Handling Costs Current Facility Layout Relocation Costs Facility Layout Design Output + Machine Locations + Material Flow Plan Figure 7 - Reconfigurable Facility Layout Methodology 17 4.4 Agile Layouts Most existing layout procedures are based on a static measure of material handling cost. However, this measure does not take into account the effect of layout on operational performance, such as cycle time, work-in-process (WIP) accumulation, queueing times at departments, and throughput rate. As production planning periods shrink, these measures become increasingly important to the performance of the agile factory where reducing manufacturing cycle times and keeping low inventory levels is key to competitiveness. Therefore, adopting layout configurations that can support the needs for low inventories, short manufacturing lead times, and high throughput rates are increasingly being pursued by industry. Unfortunately, capturing the relationship between layout configuration and operational performance has been notoriously difficult. In a recent review of over 150 papers published over the past ten years on factory layout, Meller and Gau [54] identified only one paper on the subject. This is primarily due to the lack of analytical models that are capable of explicitly capturing the effect of layout configuration on operational performance. In an initial effort [9], we developed a queueing model that allows us to capture the effect of layout configuration on key performance metrics, such as cycle time, WIP, and throughput rate. The manufacturing facility is modeled as a central server queueing network. Each processing department is modeled as either a single or a multi-server queue with general distribution of product processing and inter-arrival times. The material handling system operates as a central server in moving material from one department to another. We assume that the material handling system consists of discrete devices (e.g., forklift trucks, human operators and automated guided vehicles). The distances traveled by the material transporters are determined by the layout configuration, product routings and product demands. In determining the transporter travel time distribution, we account for both empty and full trips made by the material transport devices. Using the model, we showed that layout configuration does indeed have a direct impact on operational performance, often in unpredictable and surprising ways. For example, minimizing full travel can cause empty travel to increase, which, in turn, can increase congestion and delays. Thus it can be highly desirable to place departments in neighboring locations even though there is no direct material flow between them as this may reduce empty travel sufficiently enough to reduce overall utilization of the material handling system. This occurs, for example, if some departments are visited more frequently than others. In this case, there is a higher proportion of empty travel from and to these departments. Placing these departments in neighboring locations, although there may not be any direct flows between them, could significantly reduce empty travel. Likewise, it can be beneficial to place departments with high inter-material flows in distant locations from each other. We illustrate this by using the example layouts 18 shown in Figure 8, where we have a single product that goes through the following sequence of departments 0→1→2→3→2→2→3→4→5→6→7→8→9→8→9→10→11. It is not difficult to show that congestion, as measured by average WIP, is far worse in layout l1 than in layout l2, even though layout l1 minimizes full travel. In layout l2, departments 2, 3, 8 and 9 and 10, which are more frequently visited than other departments, are placed in adjoining locations. Despite the fact that there is no direct flows between the department pair (2, 3) and (8,9), the overall effect is a reduction in empty travel, which is sufficient to reduce the utilization of the material handling and result in an overall reduction in WIP. In general, we found that using a design criterion based on average travel distances is a poor indicator of operational performance. In fact, we can show that a layout that is optimal with respect to full travel could be operationally infeasible - i.e., it results in infinite queuing or WIP accumulation. Similarly, we can show that two layouts that are optimal with respect to full travel could have vastly different WIP values. Because conventional approaches tend to optimize average travel distance by the material handling system, they do not account for the variance in these distances. Using a queueing model, we found that distance variance plays an important role in determining how much congestion a particular layout would exhibit. More importantly, we found that congestion (for example as measured by WIP) is not necessarily monotonic in the average travel distance by the material handling system. This means that a layout which reduces average distances, but with an associated increase in variance, could lead to an overall increase in congestion. Similarly, a reduction in variance, even if it is accompanied by an increase in total travel distances, could reduce overall congestion in the system. These results point to the need to explicitly account for travel time variance when selecting a layout. A layout that exhibits a small variance may indeed be more desirable than one with a smaller average travel time. In practice, travel time variance is often dictated by the material handling system. This is especially the case for systems with automated material handling. Therefore, special attention should be devoted to identifying material handling configurations that minimize not only mean but also variance of travel distances. For example, the star-layout configuration shown in Figure 9(a) has a significantly smaller variance than the loop layout of 9(b), which itself has a smaller variance than the linear layout of 9(c). 4.4.1 Research Challenges Several avenues for future research remain to be explored. Analytical models that account for different routing and dispatching policies of the material handling system need to be constructed. These models could then be used to study the effect of different policies on layout performance. Furthermore, it will be highly valuable to use the queueing model to evaluate and compare the performance of different classical layout configurations, such as product, process, and cellular layouts, under varying conditions. This could lead to identifying new configurations that are more effective in achieving small WIP levels. 19 0 1 2 3 7 6 5 4 8 9 10 11 (a) Layout l1: uempty = 0.679, ufull = 0.311, WIP = 99.00 0 1 6 5 11 2 3 4 7 8 9 10 (a) Layout l2: uempty = 0.542, ufull = 0.409, WIP = 19.41 Figure 8 – The effect of empty travel on WIP accumulation (uempty and ufull refer to the empty and full utilization of the material handling system) (a) Uni-directional linear layout (b) Loop layout (c) Star layout Figure 9 – Material handling system configurations 20 In [9], we showed that variability plays an important role in determining WIP levels. One source of variability is that of travel times or, equivalently, travel distances. Therefore, identifying configurations that reduce distance variance, without significantly affecting average distance, can be greatly valuable. Examples of such layouts, could include a star layout configuration, where departments are equi-distant from each other, or a hub-and-spoke layout, where each hub consists of several equi-distant departments and is serviced by a dedicated transporter. In many applications, it is useful to differentiate between WIP at different departments and/or different stages of the production process. In fact, the value of WIP tends to appreciate as more work is completed and more value is added to the product. Therefore, it is more meaningful to assign different holding costs for WIP at different stages. This means that we would then favor layouts that reduce the most expensive WIP first. This can be achieved, for example, by letting departments that participate in the last production steps be as centrally located as possible. Another important avenue of research is to integrate layout design with the design of the material handling system. For example, we could simultaneously decide on material handling capacity (e.g., number of transporters) and department placement, with the objective of minimizing both WIP holding cost and capital investment costs. This would allow us to more effectively examine the tradeoffs between capacity and WIP. 5. Some Emerging Trends in Industry In this section, we report on some emerging trends in industry that could affect layout design in the future. Some of these trends support the layout configurations we discussed in the previous section. Others could transform the layout design problem in significantly different ways or even eliminate it. Our selection of trends is not meant to be exhaustive. We use it to simply highlight the potential interaction between new business practices, new technologies, and layout design. Contract Manufacturing – In many industries, outside suppliers are increasingly carrying out most of product manufacturing and assembly for the original equipment manufacturers (OEM’s) [25, 53]. Coupled with just-in-time deliveries, this has led to a reconfiguration of final assembly facilities to accommodate the closer coupling between suppliers and OEM’s. For example, many of the automotive assembly plants are allowing suppliers to deliver components directly to the point-of-use on the final assembly line. This has meant designing multiple loading docks and multiple inventory drop-off points throughout the factory - a good example is the new Cadillac plant in Lansing, Michigan which has been T-shaped to maximize supplier access to the factory floor [26]. Some automobile manufacturers, such as Volkswagen (VW), are taking this a step further by allowing suppliers to carry out some or all of the manufacturing and assembly on site [71]. The new VW truck plant in Resende, Brazil is a showcase for 20 this so-called modular plant concept [71]. To support modular plants, factories are being designed using the spine layout concept (see Figure 10), where the product moves along a main artery, or spine, through the plant. Linked to the spine are mini-assembly lines owned by the suppliers, each attaching its own module to the moving product. The layout has the hybrid features of both a flow line and multiple autonomous cells. The configuration allows the addition and removal of suppliers without affecting the main layout. It also accommodates gracefully both growth and contraction of supplier operations. Trotter, Inc., a leading manufacturer of high-end exercise treadmills, has employed similar ideas in its plant [19]. Others companies, such as General Motors, are opting instead for co-locating suppliers in a single large complex [79]. The GM Gravatai plant, also in Brazil, houses a final assembly plant and 16 supplier plants, including plants owned by Delphi, Lear, and Goodyear. Their job is to deliver pre-assembled modules to GM's line workers, who then piece the cars together in record time. The 17 plants are within walking distance from each other and are connected through a shared material handling system of forklift trucks and conveyors. The challenge in this case is to design a layout for each of the individual plants so that material handling throughout the complex is efficient and not to focus only on the local optimization of each plant. SPINE OEM’s assembly line Supplier’s production line Figure 10 - Spine layout for modular plants Delayed Product Differentiation – Increased product variety and the need for mass customization has led many companies to adopt a strategy of delayed product differentiation [19, 46, 47]. By using delayed differentiation, the point in the manufacturing process in which products are assigned individual features is postponed as much as possible. This is accomplished, for example, by building a platform common to all products. The platform is differentiated only after demand is realized by assigning to it certain product-specific features and components. Implementing delayed differentiation creates a hybrid facility consisting of a flow line-like component, where the common platform is built, and a job shop-like component where customization takes place. In the case where final products can be easily grouped into 21 families, the job shop structure could be replaced by a set of cells, each dedicated to one of the product families (see figure 11). Hence, delayed differentiation could give rise to structures similar to those discussed in section 4.2 and could benefit from the associated design tools. However, if taken to the extreme, delayed differentiation could also make the layout design problem obsolete. For example, if the customization step is taken out of the factory and is carried out at the point of sale or in distribution warehouses, as it is increasingly the case in the computer industry [51], factory design would reduce to that of a single high volume/low variety production line. Undifferentiated product stage (make-to-stock production) Product customization stage (make-to-order production) Figure 11 - Hybrid layout for plants with delayed differentiation Multi-Channel Manufacturing – The increased emphasis on quick response manufacturing, coupled with the difficulty of maintaining finished stock inventory (due to demand unpredictability or high product variety) has led many manufacturers and suppliers to invest in additional capacity, often in the form of parallel production lines [20, 53, 55]. The goal from acquiring excess capacity is to reduce cycle time by minimizing the time products spend in queueing. To take full advantage of the additional capacity, these production lines are often shared across multiple products. Thus, depending on downstream congestion, each product can move in and out of a production line to be processed on neighboring lines. EFTC, a leading contract manufacturer for electronic goods and components, offers a good example of multi-channel manufacturing [53]. A company executive of EFTC describes the production process as “small production lots moving to any of the standardized production points on the parallel production lines, passing from one line to wherever it is necessary to break bottlenecks and keep products rolling.” The concept of shared parallel production lines has also been successfully used at Sun 22 Microsystems for its line of desktop workstations [20]. The Sun manufacturing facility consists of three identical lines or cells. Each cell, in turn, has two mirror image sides which can be turned on or off, giving Sun up to six parallel production lines. As long as flow patterns and product routings do not change significantly, setting up parallel and linear production lines, similar to those at EFTC or SUN, would indeed provide the necessary flexibility and cycle time reduction. However, a linear structure becomes inefficient when operation sequences vary from product to product. An alternative might be a concentric configuration consisting of multiple identical loops, which would retain the benefits of parallel lines while accommodating a wider variety of routing. In order to reduce material handling effort, production would be assigned to inner loops, with flows venturing to outer loops only when congestion arises. A drawback of concentric layouts is the potential increase in space requirement. Scalable Machines – In the last few years, there has been a concerted effort in the metal cutting industry to develop machines that are highly flexible and scalable. These machines are to be multifunctional and capacity adjustable. This means that the basic functionality and efficiency of the machines can be easily upgraded by plugging in additional modules or acquiring additional software. Such an effort is being led by the multi-national Initiative on Intelligent Manufacturing Systems (http://www.ims.org), and supported by a large conglomerate of Japanese, US and European machine tool makers [36]. A parallel effort is also being carried out by the National Science Foundation Engineering Research Center on Reconfigurable Machines at the University of Michigan (http://erc.engin.umich.edu/), where the focus is on building easily customizable machines that match the needs of a changing product mix and production volumes [45]. If successful, these and other efforts could lead to manufacturing facilities where most of the processing takes place on only one machine, making material handling and material movement minimal. In turn, this could make layout design a less critical function for factory planning. An example of a commercial product that already exhibits some of these capabilities is the TRIFLEX machining center, marketed by Turmatic Systems. The center allows simultaneous machining using up to 7 machining units and retrofitting of additional machining units. Automatic loading and unloading systems can be easily fitted with potential for full integration into equal or other machine systems. Especially significant is the fact that a single machining unit can be fitted to a long base slide enabling the machining of all sides of a workpiece in one station and machining of the front face in another station. Therefore, 5-side machining is possible, even with only 2 machining units fitted. Portable Machines – Several equipment manufacturers are beginning to market portable machines that can be easily and dynamically deployed and re-deployed in different areas of the factory in response to changing production requirements. We mention two such examples from the machine tool industry. The TRAK QuikCell QCM-1 available from Southwestern Industries, Inc. (www.southwesternindustries.com) 23 is a compact and mobile milling machine that has found application in small lot, job shop machining. The machine can be located in close proximity to the one or two primary machining and/or turning centers dedicated to the production of a family of parts that require preliminary or secondary operations on other machines. The foundation of the machine tool consists of a base casting for easy moving with a pallet jack from any side. The small footprint of the machine allows it to fit through most doors and its rigid frame (2750 lbs.) does not require re-leveling after moving. Quick disconnects are available for electrical supply, air for coolant sprayer, power draw bar and air hose. The second example of a company that produces portable machine tools is Climax Portable Machine Tools, Inc. (www.cpmt.com). Their portable machine tools has the same functionality of stationary machine tools used for repairing turbines, paper machinery, heavy equipment, etc. – the portable machine tool goes to the workplace and it mounts on the workpiece – instead of the other way around. In this case, workpiece is stationary and movement is incurred by the machine. Hence, factory layout would have to be designed to facilitate the flow of machines instead that of parts. In Northern Telecom’s facility in Calgary, Canada which manufactures business telephone equipment, generic, modular, conveyor-mounted work cells can be easily moved from one location to another with minimum downtime [18]. Due to the relative independence of these cells, they can be unplugged from the main assembly line and moved to a new location depending upon the current product being assembled. Since this facility faces a very high frequency of product design changes, the conveyormounted work cells allow tooling and layout to be changed just before and to suit the new production and assembly schedule. With portable machine tools, the issue of machine storage and retrieval becomes important. Fortunately, there is technology being developed that allows the easy storage and retrieval of large equipment. For example, Robotic Parking Inc.’s (www.roboticparking.com/tech.html) has developed a Modular Automated Parking System (MAPS) that integrates computer control with mechanical lifts, pallets and carriers to park and retrieve large equipment in multi-level modular warehouses. Complete facilities can be constructed with lots as small as 60 ft. by 60 ft., up to 20-stories, and above ground or underground. Although originally designed for building automated parking garages, the technology is finding applications in manufacturing and warehousing. For systems with portable machines, the machines could be maintained in a MAPS-like warehouse adjacent to the main manufacturing floor. Depending on the product mix and demand volumes, machine tools would be “picked from the shelves” and co-located into the manufacturing line. 24 6. Concluding Remarks In this article, we provided an overview of emerging trends in design of next generation factory layouts. We surveyed existing academic literature dealing with design of layouts in dynamic environments. We highlighted some of the limitations of current practice in layout design and outlined some challenges that need to be addressed by both the research community and practitioners. To this effect, we described research being carried out by a recently formed university consortium on Next Generation Factory Layouts whose goal is to address some of these challenges. Finally, we reported on some emerging trends in industry that could affect layout design in the future. The goals of this article are to stimulate thought and discussion about alternative and novel ways of organizing factories of the future. We hope we are in some small measure successful in provoking thought and laying out possibilities for future research directions. Acknowledgments: This research is, in part, funded by the National Science Foundation under grants DMII 9908437, 9900039, and 9821033. Additional support has been provided by Polarfab, St Jude Medical, Recovery Engineering, Motorola, Honeywell, HP, and the St Onge Corporation. 25 References [1] Ang, C. L. and P. C. T. Willey, “A Comparative Study of the Performance of Pure and Hybrid Group Technology Manufacturing Systems Using Computer Simulation Techniques,” International Journal of Production Research, 22, 2, 193-233, 1984. [2] Arimond, J., and Ayles, W.R., “Phenolics Creep Up on Engine Applications,” Advanced Materials and Processes, 143, 20, 1993. [3] Asari, M., “Electron Beam Hardening System,” Advanced Materials and Processes, 143, 30-31, 1993. [4] Askin, R. G., N. H. Lundgren and F. Ciarallo, “A Material Flow Based Evaluation of Layout Alternatives for Agile Manufacturing,” In R. J. Graves, L. F. McGinnis, D. J. Medeiros, R. E. Ward and M. R. Wilhelm (Eds.), Progress in Material Handling Research, Braun-Brumfield, Inc., Ann Arbor, MI, 71-90, 1997. [5] Balakrishnan, J., “The Dynamics of Plant Layout,” Management Science, 39, 5, 654-655, 1993. [6] Balakrishnan, J., and Cheng, C.H., “Dynamic Layout Algorithms: A State of the Art Survey,” Omega, 26, 4, 507-521, 1998. [7] Balakrishnan, J., Jacobs, F.R., and Venkataramanan, M.A., “Solutions for the Constrained Dynamic Facility Layout Problem,” European Journal of Operational Research, 57, 280-286, 1992. [8] Batta, R., “The Dynamics of Plant Layout,” Management Science, 33, 1065, 1987. [9] Benjaafar, S., “Design of Plant Layouts with Queueing Effects,” Management Science, in press. [10] Benjaafar, S., “Flexible Factory Layouts,” Progress in Material Handling Research, Material Handling Institute, R. Graves et al. (Editors), Material Handling Institute, Charlotte, NC, 2000. [11] Benjaafar, S., “Machine Sharing in Cellular Manufacturing Systems,” Planning, Design, and Analysis of Cellular Manufacturing Systems, A. K. Kamrani, H. R. Parasei and D. H. Liles (Editors), Elsevier Science B. V., 1995. [12] Benjaafar, S. and D. Gupta, “Scope versus Focus: Issues of Flexibility, Capacity, and Number of Production Facilities,” IIE Transactions, 30, 5, 413-425, 1998. [13] Benjaafar, S, and M. Sheikhzadeh, “Design of Flexible Plant Layouts,” IIE Transactions, 32, 309322, 2000. [14] Bullington, S. F. and D. B. Webster, “Evaluating the Flexibility of Facilities Layouts Using Estimated Relayout Costs,” Proceedings of the IXth International Conference on Production Research, 2230-2236, 1987. [15] Conway, D.G., and Venkataramanan, M.A., “Genetic Search and the Dynamic Facility Layout Problem,” Computers and Operations Research, 21, 8, 955-960, 1994. [16] Drolet, J. R., “Scheduling Virtual Cellular Manufacturing Systems,” Ph.D. Thesis, School of Industrial Engineering, Purdue University, West Lafayette, Indiana, 1989. 26 [17] Editor, “Technology Trends: Concepts for Work and Tool-holding,” American Machinist, 137, 10, 1993. [18] Editor, “Norstart Custom Telephones Rely on Flexible Conveyor Systems,” Assembly Magazine Online, May 1996. [19] Editor, “Flexible Workstations Cut Work-In Process,” Assembly Magazine Online, September 1995. [20] Feare, T., “Less Automation Means More Productivity at Sun Microsystems,” Modern Materials Handling, November 1, 1997. [21] Feitzinger, E. and H. Lee, “Mass Customization at Hewlett Packard: The Power of Postponement,” Harvard Business Review, January-February, 116-121, 1997. [22] Flynn, B. B., and Jacobs, F. R. “A Simulation Comparison of Group Technology with Traditional Job Shop Manufacturing,” International Journal of Production Research, 24, 1171-1192, 1986. [23] Fu, M. C. and B. K. Kaku, “Minimizing Work-in-Process and Material Handling in the Facilities Layout Problem,” IIE Transactions, 29, 1-29, 1997. [24] Fujine, M., Kaneko, T., and Okijima, J. “Aluminum Composite Replace Cast Iron,” Advanced Materials and Processes, 143, 34-36, 1993. [25] Gibson, P., “The Asset Paradox,” Electronic Business Magazine, April, 2000. [26] Green, J., “Why Workers Are Lining Up for Jobs at This GM Plant,” Business Week, October 2, 2000. [27] Gupta, R. M., “Flexibility in Layouts: A Simulation Approach,” Material Flow, 3, 243-250, 1986. [28] Hicks, P.E., and Cowan, T.E., “CRAFT-M for Layout Arrangement,” Industrial Engineering, 8, 3035, 1976. [29] Heragu, S. S., Facilities Design, PWS Publishing Co., Boston, MA, 1997. [30] Heragu, S. S., “Group Technology and Cellular Manufacturing,” IEEE Transactions on Systems, Man and Cybernetics, 24, 2, 203-215, 1994. [31] Heragu, S. S. and Kochhar, J.S., “Material Handling Issues in Adaptive Manufacturing Systems”, The Materials Handling Engineering Division 75th Anniversary Commemorative Volume, ASME, New York, NY, 1994. [32] Heragu, S. S. and Zijm, W.H.M., “A Framework for Designing Reconfigurable Facilities,” Technical Report, DSES Department, Rensselaer Polytechnic Institute, Troy, NY 12180, 2000. [33] Heragu, S. S., Zijm, W.H.M., van Ommeren, J.C.W. and Meng, G., “An Open Queuing Network Approach to Evaluating Cellular and Jobshop Layouts,” Technical Report, DSES Department, Rensselaer Polytechnic Institute, Troy, NY 12180, 2000. 27 [34] Huang, H., and Irani, S. A., “Facility Layout Using Operation Sequences – Part 2: A New Similarity Measure for Operation Sequences”. Working Paper, Department of Industrial, Welding and Systems Engineering, The Ohio State University, Columbus, OH, 1999. [35] Huang, H., and Irani, S.A., “Design of Facility Layouts using Layout Modules: A Numerical Clustering Approach”. Proceedings of the 8th Industrial Engineering Research Conference, May 2326, Phoenix, AZ, 1999. [36] Ikegaya, A. “Highly Productive And Reconfigurable Manufacturing System,” Technical Project Report, Intelligent Manufacturing Systems Initiative, www.ims.org, 2000. [37] Irani, S. A., and Huang, H., “Custom Design of Facility Layouts for Multi-Product Facilities using Layout Modules,” IEEE Transactions on Robotics and Automation, 16, 259-267, 2000. [38] Irani, S. A., and Huang, H., “Layout Modules: A Novel Extension of Hybrid Cellular Layouts,” Proceedings of the 1998 International Mechanical Engineering Congress & Exposition, Winter Annual Meeting of the ASME, November 15-20, Anaheim, CA, 1998. [39] Irani, S.A., “Production Flow Analysis in a Semiconductor Fab,” Proceedings of the Joint National Science Foundation and Semiconductor Research Corporation Workshop on Operational Methods in Semiconductor Manufacturing, University of California at Berkeley, CA, February 19-20, 1997. [40] Irani, S. A., T. M. Cavalier, and P. H. Cohen, “Virtual Manufacturing Cells: Exploiting Layout Design and Intercell Flows for the Machine Sharing Problem,” International Journal of Production Research, 31, 791-810, 1993. [41] Kaku, B.K., and Mazzola, J. B., “A Tabu-Search Heuristic for the Dynamic Plant Layout Problem,” INFORMS Journal on Computing, 9, 374-384, 1997. [42] Kochhar, J.S. and S.S. Heragu, “Facility Layout Design in a Changing Environment,” International Journal of Production Research, 37, 2429-2446, 1999. [43] Kochhar, J.S. and Heragu, S.S., “MULTI-HOPE: A Tool for Multiple Floor Layout Problems,” International Journal of Production Research, 36, 3421-3435, 1998. [44] Kochhar, J.S., Heragu, S.S. and Foster, B., “HOPE: A Genetic Algorithm for Unequal Area Facility Layout Problem,” Computers and Operations Research, 25, 583-594, 1998. [45] Koren, Y., Jovane, F., Moriwaki, T., Pristchow, G., Ulsoy, G. & Brusell H.V., “Reconfigurable Manufacturing Systems,” Annals of the CIRP, 48, 527-539, 1999. [46] Kouvelis, P., and Kiran, A.S., “Single and Multiple Period Layout Models for Automated Manufacturing Systems,” European Journal of Operational Research, 52, 300-314, 1991. [47] Kouvelis, P., Kurawarwala, A.A., and Gutierrez, G.J., “Algorithms for Robust Single Period and Multiple Period Layout Planning for Manufacturing Systems,” European Journal of Operational Research, 63, 287-303, 1992. [48] Kusiak, A. and S. S. Heragu, “The Facility Layout Problem,” European Journal of Operational Research, 27, 229-251, 1987. 28 [49] Lahmar, M. and S. Benjaafar, “Design of Plant Layouts for Expansion,” Working Paper, Department of Mechanical Engineering, University of Minnesota, Minneapolis, 2000. [50] Lacksonen, T.A., and Enscore, E.E., “Quadratic Assignment Algorithms for the Dynamic Layout Problem,” International Journal of Production Research, 31, 3, 503-517, 1993. [51] Lee, H. L., and C. S. Tang, “Modeling the Costs and Benefits of Delayed Product Differentiation,” Management Science, 43, 40-53, 1997. [52] Magretta, J., “The Power of Virtual Integration: An Interview with Dell Computer's Michael Dell,” Harvard Business Review, March-April (1998), 73-84. [53] McHale, T., “Special Report - The Top 100 Contract Manufacturers,” Electronic Business, August, 1999. [54] Meller, R. and K. Y. Gau, “The Facility Layout Problem: Recent and Emerging Trends and Perspectives,” Journal of Manufacturing Systems, 15, 351-366, 1996. [55] Meller, R. and R. L. DeShazo, “Multi-Channel Manufacturing at Electrical Box & Enclosures,” Working Paper, Department of Industrial and Systems Engineering, Virginia Tech, 2000. [56] Montreuil, B. and LaForge, A., “Dynamic Layout Design Given a Scenario Tree of Probable Futures,” European Journal of Operational Research, 63, 271-286, 1992. [57] Montreuil, B. and Venkatadri, U., "Strategic Interpolative Design of Dynamic Manufacturing Systems Layout," Management Science, 37, 682-694, 1991. [58] Montreuil, B., Venkatadri, U. and P. Lefrançois, “Holographic Layout of Manufacturing Systems,” Technical Report No. 91-76, Faculty of Management, Laval University, Québec, Canada, 1991. [59] Montreuil B., Venkatadri. U. and R.L. Rardin, “The Fractal layout Organization for Job Shop Environments,” International Journal of Production Research, 1999. [60] National Research Council, Visionary Manufacturing Challenges for 2020, National Academy Press, Washington, D.C., 1998. [61] Norman, B. A. and A. E. Smith, “Considering Production Uncertainty in Block Layout Design,” Working Paper, Department of Industrial Engineering, University of Pittsburgh, Pittsburgh, PA, 2000. [62] Palekar, U.S., Batta, R., Bosch, R.M., and Elhence, S., “Modeling Uncertainties in Plant Layout Problems,” European Journal of Operational Research, 63, 347-359, 1992. [63] Rosenblatt, M. J., “The Dynamics of Plant Layout,” Management Science, 32, 76-86, 1986. [64] Rosenblatt, M. J., and Kropp, D. H., “The Single Period Stochastic Plant layout Problem,” IIE Transactions, 24, 169-76, 1992. [65] Rosenblatt, M. J. and Lee, H.L., “A Robustness Approach to Facilities Design,” International Journal on Production Research, 25, 479-486, 1987. 29 [66] Sarper, H. and T. J. Greene, “Comparison of Equivalent Pure Cellular and Functional Production Environments Using Simulation,” International Journal of Computer Integrated Manufacturing, 6, 221-236, 1993. [67] Sethi, A. K. and S. P. Sethi, 1990, “Flexibility in Manufacturing: A survey,” The International Journal of Flexible Manufacturing Systems, 2, 289-328. [68] Shafer, S. M. and J. M. Charnes, “Cellular Versus Functional Layout under a Variety of Shop Operating Conditions,” Decision Sciences, 36, 333-342, 1988. [69] Shore, R. H. and J. A. Tompkins, “Flexible Facilities Design,” AIIE Transactions, 12, 200-205, 1980. [70] Suresh, S. C., “Partitioning Work Centers for Group Technology: Analytical Extension and ShopLevel Simulation Investigation,” Decision Sciences, 23, 267-290, 1992. [71] Smith, G., Wheatley, J. and J. Green, “Car Power,” Business Week – International Edition, October 23, 2000. [72] Tompkins, J. A., “Modularity and Flexibility: Dealing with Future Shock in Facilities Design,” Industrial Engineering,, September, 78-81 1980. [73] Tompkins, J. A., J. A. White, Y. A. Bozer, E. H. Frazelle, J. M. A. Tanchoco, and J. Trevino, Facilities Planning, 2/E, John Wiley, New York, NY, 1996. [74] Urban, T.L., “Solution Procedures for the Dynamic Facility Layout Problem,” Annals of Operations Research, 76, 323 – 342, 1998. [75] Urban, T.L., “A Heuristic for the Dynamic Layout Problem” IIE Transactions, 25, 4, 57-63, 1993. [76] Urban, T.L., “Computational Performance and Efficiency of Lower Bound Procedures for the Dynamic Facility Layout Problem,” European Journal of Operational Research, 57, 271–279, 1992. [77] Venkatadri, U., Rardin, R. and B. Montreuil, “A Design Methodology for the Fractal Layout Organization,” IIE Transactions, 29, 911-924, 1997. [78] Venkatadri, U., R. L. Rardin and B. Montreuil, “Facility Organization and Layout Design: An Experimental Comparison for Job Shops,” Technical Report No. 96-27, Faculty of Management, Laval University, Québec, Canada, 1996. [79] Wheatley, J. “Super Factory – or Super headache,” Business Week, July 31, 2000. [80] Webster, D. B. and M. B. Tyberghein, “Measuring Flexibility of Job Shop layouts,” International Journal of Production Research, 18, 21-29, 1980. [81] Wemmerlöv, U. and L. N. Hyer, “Cellular Manufacturing in the U.S. Industry: A Survey of Users,” International Journal of Production Research, 27, 1511-1530, 1989. [82] Yang, T., and Peters, B.A., “Flexible Machine Layout Design for Dynamic and Uncertain Production Environments,” European Journal of Operational Research, 108, 49-64, 1998. 30