MASS CUSTOMIZATION:MASS CUSTOMIZATION MANUFACTURING

MASS CUSTOMIZATION MANUFACTURING

Competition for mass customization manufacturing is focused on the flexibility and responsiveness in order to satisfy dynamic changes of global markets. The traditional metrics of cost and quality are still necessary conditions for companies to outpace their competitors, but they are no longer the deciding factors between winners and losers. Major trends are:

1. A major part of manufacturing will gradually shift from mass production to the manufacturing of semicustomized or customized products to meet increasingly diverse demands.

2. The ‘‘made-in-house’’ mindset will gradually shift to distributed locations, and various entities will team up with others to utilize special capabilities at different locations to speed up product development, reduce risk, and penetrate local markets.

3. Centralized control of various entities with different objectives, locations, and cultures is almost out of the question now. Control systems to enable effective coordination among distributed entities have become critical to modern manufacturing systems.

To achieve this, it is becoming increasingly important to develop production planning and control architectures that are modifiable, extensible, reconfigurable, adaptable, and fault tolerant. Flexible manufacturing focuses on batch production environments using multipurpose programmable work cells, automated transport, improved material handling, operation and resource scheduling, and com- puterized control to enhance throughput. Mass customization introduces multiple dimensions, in- cluding drastic increase of variety, multiple product types manufactured simultaneously in small batches, product mixes that change dynamically to accommodate random arrival of orders and wide spread of due dates, and throughput that is minimally affected by transient disruptions in manufac- turing processes, such as breakdown of individual workstations.

Managing Variety in Production Planning

Major challenge of mass customization production planning results from the increase of variety. The consequence of variety may manifest itself through several ramifications, including increasing costs due to the exponential growth of complexity, inhibiting benefits from economy of scale, and exac- erbating difficulties in coordinating product life cycles. Facing such a variety dilemma, many com- panies try to satisfy demands from their customers through engineering-to-order, produce-to-order, or assembly-to-order production systems (Erens and Hegge 1994).

At the back end of product realization, especially at the component level and on the fabrication aspect, today we have both flexibility and agility provided by advanced manufacturing machinery such as CNC machines. These facilities accommodate technical variety originating from diverse needs of customers. However, at the front end, from customer needs to product engineering and production planning, managing variety is still very ad hoc. For example, production control information systems, such as MRPII (manufacturing resource planning) and ERP (enterprise resource planning), are falling behind even though they are important ingredients in production management (Erens et al. 1994). The difficulties lie in the necessity to specify all the possible variants of each product and in the fact that current production management systems are often designed to support a production that is based on a limited number of product variants (van Veen 1992).

The traditional approach to variant handling is to treat every variant as a separate product by specifying a unique BOM for each variant. This works with a low number of variants, but not when customers are granted a high degree of freedom in specifying products. The problem is that a large number of BOM structures will occur in mass customization production, in which a wide range of combinations of product features may result in millions of variants for a single product. Design and maintenance of such a large number of complex data structures are difficult, if not impossible. To overcome these limitations, a generic BOM (GBOM) concept has been developed (Hegge and Wort- mann 1991; van Veen 1992). The GBOM provides a means of describing, with a limited amount of data, a large number of variants within a product family, while leaving the product structure unim- paired. Underlying the GBOM is a generic variety structure for characterizing variety, as schemati- cally illustrated in Figure 8. This structure has three aspects:

1. Product structure: All product variants of a family share a common structure, which can be described as a hierarchy containing constituent items (Ii) at different levels of abstraction, where

{Ii} can be either abstract or physical entities. Such a breakdown structure (AND tree) of {Ii} reveals the topology for end-product configuration (Suh 1997). Different sets of Ii and their interrelationships (in the form of a decomposition hierarchy) distinguish different common product structures and thus different product families.

2. Variety parameters: Usually there is a set of attributes, A, associated with each Ii. Among them, some variables are relevant to variety and thus are defined as variety parameters,

{Pj} C A. Like attribute variables, parameters can be inherited by child node(s) from a parent node. Different instances of a particular Pj, e.g., {Vk}, embody the diversity resembled by, and perceived from, product variants.

Two types of class-member relationships can be observed between {Pj} and {Vk}. A leaf Pj (e.g., P32) indicates a binary-type instantiation, meaning whether I32 is included in I3 (V32 = 1), or not (V32 = 0). On the other hand, a node Pj (e.g., P2) indicates a selective type instantiation, that is, I2 has several variants in terms of values of P2, i.e., V2 � {V2 1, V2 2}.

3. Configuration constraints: Two types of constraint can be identified. Within a particular view of product families, such as the functional, behavioral, or physical view, restrictions on the

combination of parameter values, {Vk}, are categorized as Type I constraints. For example, V11-1 1 and V31-3 2 are incompatible, that is, only one of them can be selected for a product variant, indicating an exclusive all (XOR) relationship. The mapping relationships of items and their variety parameters across the functional, behavioral, and structural views are referred to as Type II constraints. This type of constraint deals mostly with configuration design knowl- edge. Usually they are described as rules instead of being graphically depicted in a generic structure. While the functional and behavioral views of product families are usually associated with product family design, the major concern of managing variety in production is Type I constraints which mostly involves the structural view.

Table 2 illustrates the above generic variety structure using a souvenir clock example. As far as variant handling is concerned, the rationale of the generic variety structure lies in the recognition of the origin and subsequent propagation of variety. Three levels of variation have been indicated, that is, at the structure, variety parameter, and instance levels. Different variation levels have different variety implications. To understand the ‘‘generic’’ concept underlying such a variety representation, two fundamental issues need to be highlighted:

1. Generic item: A generic item represents a set of similar items (called variants) of the same type (a family). The item may be an end product, a subassembly, an intermediate part, or a component part (van Veen, 1992). It may also be a goes-into-relationship or an operation. For example, a red front plate (I 1*), a blue front plate (I 2*) and a transparent front plate (I 3*) are three individual variants, whereas a generic item, I, represents such a set of variants (a family of front plates), that is I � {I *1 , I 2*, I *3 }. However, these variants are similar in that they share a common structure (e.g., BOM structure) in configuring desk clocks.

2. Indirect identification: Instead of using part numbers (referred to as direct identification), the identification of individual variants from a generic item (within a family) is based on variety parameters and their instances (a list if parameter values). Such identification is called indirect identification (Hegge and Wortmann. 1991). In the above example, a variety parameter, color, and its value list, , can be used for an indirect identification of a particular variant,

Coordination in Manufacturing Resource Allocation

Mass customization manufacturing is characterized by shortened product life cycle with high-mixed and low-volume products in a rapidly changing environment. In customer-oriented plants, orders consisting of a few parts, or even one part, will be transferred directly from vendors to producers, who must respond quickly to meet short due dates. In contrast to mass production, where the man- ufacturer tells consumers what they can buy, mass customization is driven by customers telling the manufacturer what to manufacture. As a result, it is difficult to use traditional finite capacity sched- uling tools to support the new style of manufacturing. Challenges of manufacturing resource allo- cation for mass customization include:

1. The number of product variety flowing through the manufacturing system is approaching an astronomical scale.

2. Production forecasting for each line item and its patterns is not often available.

3. Systems must be capable of rapid response to market fluctuation.

4. The system should be easy for reconfiguration—ideally, one set of codes employed across different agents.

5. The addition and removal of resources or jobs can be done with little change of scheduling systems.

Extensive research on coordination of resource allocation has been published in connection with scenarios of multiple resource providers and consumers. In the research, existence of demand patterns is the prerequisite for deciding which algorithm is applicable. The selection of a certain algorithm is often left to empirical judgment, which does not alleviate difficulties in balancing utilization and level of services (e.g., meeting due dates).

As early as 1985, Hatvany (1985) pointed out that the rigidity of traditional hierarchical structures limited the dynamic performance of systems. He suggested a heterachical system, which is described as the fragmentation of a system into small, completely autonomous units. Each unit pursues its own goals according to common sets of laws, and thus the system possesses high modularity, modifiability, and extendibility. Following this idea, agent-based manufacturing (Sikora and Shaw 1997) and holonic manufacturing (Gou et al. 1994), in which all components are represented as different agents and holons, respectively, are proposed to improve the dynamics of operational organizations.

From traditional manufacturing perspectives, mass customization seems chaotic due to its large variety, small batch sizes, random arrival orders, and wide span of due dates. It is manageable, however, owing to some favorable traits of modern manufacturing systems, such as inherent flexibility in resources (e.g., increasing use of machining centers and flexible assembly workstations) and sim- ilarities among tools, production plans, and product designs. The challenge is thus how to encode these characteristics into self-coordinating agents so that the invisible hand, in the sense of Adam Smith’s market mechanism (Clearwater 1996), will function effectively.

Market-like mechanisms have been considered as an appealing approach for dealing with the coordination of resource allocation among multiple providers and consumers of resources in a dis- tributed system (Baker 1991; Malone et al. 1988; Markus and Monostori 1996). Research on such a distributed manufacturing resource-allocation problem can be classified into four categories: the bidding / auction approach (Shaw 1988; Upton and Barash 1991), negotiation approach (Lin and Solberg 1992), cooperative approach (Burke and Prosser 1991) and pricing approach (Markus and Monostori 1996).

Major considerations of scheduling for resource allocation include:

1. Decompose large, complex scheduling problems into smaller, disjointed allocation problems.

2. Decentralize resource access, allocation, and control mechanisms.

3. Design a reliable, fault-tolerant, and robust allocation mechanism.

4. Design scalable architectures for resource access in a complex system and provide a plug-and- play resource environment such that resource providers and consumers can enter or depart from the market freely.

5. Provide guarantees to customers and applications on performance criteria.

In this regard, the agent-based, market-like mechanism suggests itself as a means of decentralized, scalable, and robust coordination for resource allocation in a dynamic environment (Tseng et al. 1997). In such a collaborative scheduling system, each workstation is considered as an autonomous agent seeking the best return. The individual work order is considered as a job agent that vies for

the lowest cost for resource consumption. System scheduling and control are integrated as an auction- based bidding process with a price mechanism that rewards product similarity and response to cus- tomer needs. A typical market model consists of agents, a bulletin board, a market system clock, the market operating protocol, the bidding mechanism, pricing policy, and the commitment mechanism. Figure 9 illustrates how the market operating protocol defines the rules for synchronization among agents.

The satisfaction of multiple criteria, such as costs and responsiveness, cannot be achieved using solely a set of dispatching rules. A price mechanism should be constructed to serve as an invisible hand to guide the coordination in balancing diverse requirements and maximizing performance in a dynamic environment. It is based on market-oriented programming for distributed computation (Adelsberger and Conen 1995). The economic perspective on decentralized decision making has several advantages. It overcomes the narrow view of dispatching rules, responds to current market needs, uses maximal net present value as the objective, and coordinates agents’ activities with minimal communication. In collaborative scheduling, objectives of the job agent are transformed into a set of evaluation functions. The weights of the functions can be adjusted dynamically on basis of system states and external conditions. Resource agents adjust the charging prices based on their capability and utilization and the state of current system. Mutual selection and mutual agreement are made through two-way communication. Figure 10 depicts the market price mechanism. In the market model, the job agents change routings (i.e., select different resource agents), and adjust Job Price as a pricing tool to balance the cost of resources and schedule exposure. Resource agents adjust their prices according to market demands on their capability and optimal utilization and returns. For example, a powerful machine may attract many job agents, and thus the queue will build up and

waiting time will increase. When a resource agent increases its price, the objective of bidding job will decrease, thus driving jobs to other resources and diminishing the demand for this resource.

Specifically, the price mechanism can be expressed as follows:

1. The job agent calculates the price to be paid for the i operation:

where s represents the setup cost, tp denotes the processing time of the operation, and PFAindex represents the product family consideration in setting up the manufacturing system. For in- stance, we can let PFAindex = 0, if two consecutive jobs, i and j, are in the same prod- uct family, and hence the setup charge can be eliminated in the following job. Otherwise, PFAindex = 1. The Opportunity cost in Eq. (3) represents the cost of losing particular slack for other job agents due to the assignment of resource for one job agent’s operation i, as expressed below:

Based on the above formulations, the collaborative control can be modeled as a message-based simulation, as shown in Figure 11. The control process is driven by an event and / or an abnormal event, such as machine breakdown or a new due date. All events can be represented as messages.

High-Variety Shop-Floor Control

Mass customization manufacturing motivates a new generation of shop-floor control systems that can dynamically respond to customer orders and unanticipated changes in the production environment. The requirements of the new control systems include reconfigurability, decomposability, and scala- bility to achieve make-to-order with very short response time. A systematic approach has been de- veloped to design control system by leveraging recent progresses in computing and communication hardware and software, including new software engineering methods and control technologies, such as smart sensors and actuators, open architectures, and fast and reliable networks (Schreyer and Tseng

1998). Figure 12 illustrates an actual example of installation of a mass customization manufacturing system in the Hong Kong University of Science and Technology.