Go to the table of contents.

Previous ...



4.6 Manufacturing system development

4.6.1 Systematic development of manufacturing systems

A manufacturing system is a technical system in itself. It should therefore be possible to apply design methods and tools used to design the product in the process of creating the manufacturing system. Several authors have investigated this approach. Among them are Aganovic et al (2002) who proposed a framework for an integrated product and process model, see Figure 36. In the context of this thesis, there are in particular two models that attract our interest. These are the product model and the manufacturing system model. In Figure 36, these two are deliverables from activities A1 and A2. During the iterative development process of the product as well as the manufacturing system these two models are referred to as open product model and open manufacturing system model. The reason for this is that they need to be known and communicated in an iterative fashion between the teams developing the product and the teams developing the manufacturing system.

A kind of roadmap for an integrated view of engineering design theories and their application in development of the product and the manufacturing system was developed by Aganovic et al. (2003), see Figure 37. A product and its manufacturing system that are subjected to design activities are each represented by a chromosome (see Theory of Domains in section 4.2.1). Their critical quantitative parameters are structured in the axiomatic design theory (ADT) domains. Unstructured qualitative customer needs (CNs) are formalized as product processes that are carried out by product functions. Product functions are solved by product organs and embodied in product components. A manufacturing process transforms product components from an original state into a desired state. A product is therefore regarded as operand in the manufacturing system. Manufacturing processes are carried out by manufacturing functions that are solved by manufacturing organs and embodied in manufacturing components. The structuring of a manufacturing process is carried out according to manufacturing system design principles about e.g. manufacturing process and workshop layout, batching and manufacturing resource resetting, manufacturing process control or inventory management. Manufacturing functions, organs and components are determined while considering various producibility principles and existing resource infrastructure at the manufacturing company. Product and manufacturing system structures created in the Theory of Domains framework are characterized by certain quantitative parameters whose values (and tolerances) must be set and properly managed. To accomplish this, quantitative parameters are introduced into the ADT framework. FRs characterize product process and product function, while product organ and product component are characterized by DPs. Manufacturing process, manufacturing functions, manufacturing organs, and manufacturing components are characterized by PVs. The property classification framework of the TTS supports the selection of FRs, DPs, and PVs. ADT is here used as an analysis framework while ToD is used as a synthesis framework. Synthesis and analysis are regarded as inseparable activities that must be executed when designing a technical system.

Figure 36: Functional model of an innovation system (adapted from Aganovic et al, 2002).

Fagerström et al. (2002) refer to the above approach as a multi-viewpoint model that is seen as a map to support navigation when working in the innovation process. The axiomatic design approach is used to develop both the product and its manufacturing system (Figure 38 a). The multi-viewpoint model (Figure 38 b) is built along two axes: the process and tollgate axis and the function, design, and iteration axis.

The order in which the main phases of a development project are executed is presented along the process and tollgate axis. There are two main phases along this axis: the development phase and the realization phase. The distinction between these two phases is clearly illustrated along the function design iteration axis. Synthesis and analysis are the main activities in the development phase, whereas make and deliver are the main activities in the realization phase.

Figure 37: The relationship between engineering design theories (Aganovic et al, 2003).

Figure 38: Multi-viewpoint model of innovation process (Fagerström et al., 2002).


4.6.2 Axiomatic design applied to manufacturing system development

Cochran (1999) presents an approach to the design and implementation of a lean production system that is based on and guided by the axiom of independence in axiomatic design (Suh, 1990). Cochran (1999) proposes a production system design framework composed of a number of elements (e.g. the production system design decomposition and deployment steps for implementation). The proposed framework communicates the elements of a multi-faceted system design in a logical and systematic way. It thereby enables communication throughout an organization. Moreover, the axiomatic design foundation on which it is based provides the scientific and theoretical basis for designing and re-designing any production system. Owing to the general foundation of the framework it is product independent in the sense that it can reflect a production system design that is applicable to any discrete-part production environment, independent of production volume and product type. Cochran et al. (2001) propose manufacturing system design decomposition (MSDD) to support the development. At the fourth level of decomposition, they organize the FR-DP pairs into six different branches (Figure 39).

Figure 39: The six different branches of a MSDD (Cochran et al., 2001).

The intention of the proposed MSDD is that it should support manufacturing system designers to: clearly separate objectives from means of achievement; relate low-level activities and decisions to high-level goals and requirements; understand inter-relationships among the different elements of a system design; and effectively communicate the information across a manufacturing organization. The MSDD enables a company to simultaneously achieve cost, quality, responsiveness in delivery to the customer, and flexibility objectives.

4.6.3 Design of assembly systems for modular products

He & Kusiak (1997) present an approach to the design of assembly systems for modular products. The special structure of modular products presents both opportunities and challenges to the manufacturing system design. He & Kusiak (1997) model the product structure using an acyclic digraph, where nodes represent operations and arcs correspond to the precedence relations between operations. Each assembly operation corresponds to a physical component in the product. An assumption is that the assembly structure of a modular product includes a basic structure and a variant structure. The problem is to optimize the line balancing and manufacturing sequence through an allocation of operations to stations on the assembly line. To solve this configuration problem they apply a heuristic algorithm to solve the configuration problem based on the tabu search approach. Tabu search is a powerful optimization procedure that has been successfully applied for solving various combinatorial optimization problems. The Tabu search approach is stated to be easy to implement and as easily incorporating problem-specific constraints. He & Kusiak (1997) describe that Tabu search as an iterative improvement procedure that begins with an initial feasible solution and that attempts to determine a better solution. Tabu is different from traditional iterative optimization algorithms in that it has the ability to escape local optima by using a short-term memory of recent solutions (tabu list). Furthermore, tabu search permits backtracking to previous solutions, which ultimately may lead – via different directions – to better solutions (aspiration). The features of a tabu list and aspiration make tabu search a powerful optimization tool for manufacturing scheduling problems. The approach presented above is further described in He & Kusiak (1998). In He et al. (1998) the graph model approach is used to analyze and design delayed product differentiation.

De Lit et al. (2003a) present a product family representation model. The purpose is to propose an integrated approach for the design of product families and their assembly system. The proposed model is able to deal with partial information from the design of the product family – a situation that often occurs at early stages of the concurrent design of products and production means. In the proposed design approach the product is decomposed into functional entities defined by the designer. The product family is seen as an assembly of functional entities. Component variants are aggregated into generic components, and links between components become generic. De Lit et al. (2003b) add to the approach described above and state that most assembly-centered concurrent engineering approaches do not deal with product families. The purpose of the research work presented is thus to develop a new philosophy for the design of a product family and its assembly system. The intention of the proposed approach is to provide the designer with a set of quick tools and a methodology to perform preliminary design and test several alternatives. Another intention of the proposed approach is that it should be possible to use the tools independently. The designer should always remain the master of the optimization process. The final step of the design is a simulation to check the validity of the constructed layout. If the results are not satisfactory, previous steps will be revisited to improve the design. In future work the authors intend to integrate the tools.

4.6.4 Flexible manufacturing systems

To provide for different product variants from a spatial as well as a generational perspective (Martin & Ishii, 1997) the manufacturing system must build some kind of flexibility and adaptability into the system. Tsubone & Horikawa (1999) present and discuss three types of flexibility regarding the performance of a production system. In this they consider machine flexibility and routing flexibility in a parts fabrication process and volume flexibility in the context of an assembly line. Machine flexibility is the ability of a machine to perform different operations required by different part variants to be processed by the machine. Routing flexibility is the ability to provide several different alternative routings for the set of part variants fabricated. Routing flexibility can be essential both in terms of dealing with different kinds of volume demands for different part variants and for maintaining through-put capabilities in terms of machine breakdowns. Volume flexibility is the ability to manage production operations at different output or working time levels.

Figure 40: Simple model of a manufacturing system (Tsubone & Horikawa, 1999)

Tsubone & Horikawa (1999) focus attention on the need to be able to evaluate different types of manufacturing flexibility that are introduced simultaneously. Since different types of manufacturing flexibility interact, there is a need for a better understanding of how manufacturing performance is affected. The authors use a simulation approach to analyze the three different types of flexibility described above in a situation where urgent orders and machine breakdowns can not be regarded as negligible. The findings presented indicate that different types of flexibility have situation dependent effects on production measurements and therefore motivate modeling and analysis in order to reach appropriate decisions. The purpose is to provide management with better guidelines for giving priority and determining the scope or scale in terms of standardization for design, process/operations improvements, and investment in equipment.

4.6.5 Petri net based manufacturing system modeling

In the view of Zha & Du (2001), an assembly is composed of parts or components and connectors (joints), and a single part is composed of physical features. The different levels of assembly form a hierarchy that utilizes the relationships between different parts of assembly and different features of part. A place-transition (P/T) model is used to represent the mechanical systems and assemblies, in which each part is represented as a place and each joint is represented as a transition. With this approach, a mechanical system or assembly can be viewed as a hierarchical P/T net and accordingly a sub-system or sub-assembly is a sub P/T net. With a modeling approach of this type, the static and dynamic characteristics in an assembly oriented design process can be captured. Zha & Du (2001) further suggest that fuzzy logic can be used to augment the model with the capability to deal with incomplete, imprecise, and uncertain knowledge. Such a hybrid design object model can incorporate a product data model, a top-down design process, and an assembly process model by using the object oriented, knowledge based, feature based, parametric and constraint based modeling approach. The model would then provide more accurate and more flexible representations for concurrent integrated assembly design and planning.

Based on the place-transition model, an assembly can be represented by the topological structure of its places (parts) and the liaison relationships (transitions and arcs) between parts. The assembly relation supplies two kinds of information: the part name that the feature links, and the type of relationship between the feature and the connected part, such as fit. Therefore, the related parts and feature links in the P/T model can be identified for every part. The system then uses a set to collect all the related parts. In the collection, the first position is filled with the part, and the remaining positions are filled with its related parts and links. Linking the collections of every part by features in a product forms the place-transition net graph of the product.

An assembly procedure is usually subjected to constraints of different parts and sub-assemblies (Zha & Du, 2001). These constraints can be broadly classified into two categories: hard constraints, and soft constraints. Hard constraints are the geometric and physical constraints related to the generation of assembly sequence. Soft constraints are a consequence of assembly planners making choices in defining an assembly sequence. Further, the constraints can be categorized as: topological constraints, geometric constraints, stability and security constraints, and partial precedence constraints. Topological constraints deal with the interconnections between parts. The topological constraint existence of a sub-assembly implies that there is at least one path in the local liaison graph from an arbitrary part in a sub-assembly to any other part in this sub-assembly. Geometric constraints of an assembly entail whether there are relative or allowable position and orientation relations between parts and collision-free paths in the assembly. Stability constraints express that and how the parts of the assembly maintain their relative positions and that they do not break contact spontaneously. Stability constraints govern the relative positions when the assembly is subjected to gravity or other forces acting upon the assembly. A secured sub-assembly has zero degree-of-freedom for relative motion of all parts in the sub-assembly. A sub-assembly without stability or security constraints means that it is unstable and changeable, and thus unfeasible. Partial precedence constraints define a part assembled with a part or a sub-assembly in a desired direction or precedence in terms of an assembly operation. Two steps are needed to find these relationships among the parts of an assembly: checking the existence of contact between each pair of parts, and checking possible disassembly directions of each part from the other. Zha & Du (2001) propose a hybrid representation model where the place-transition net model is combined with relational models and constraint models in order to support assembly planning.

Qiao et al. (2002) refer to the mass customization paradigm as a driver for the need of new approaches to modeling and simulation of manufacturing systems. The mass customization paradigm emphasizes shortened product lifecycles with the implication for a need for increased changes and re-configuration of production lines. Therefore, highly flexible and re-configurable factories must be designed, simulated, and analyzed. Qiao et al. (2002) present a methodology for representing manufacturing systems using a valid, colored Petri net. Their model is able to represent a solution to problems such as dynamic rescheduling, shop reconfiguration, part rework processing, and mechanisms for recovery from machine failure. An additional consideration is that mass customization manufacturing systems must often support product design modifications at late stages of production, and must respond and adjust to these changes quickly, without postponing delivery time. Petri net (PN) is a methodology that can be used to design discrete-event system models graphically and mathematically. A simple PN is defined as a bipartite graph consisting of places, transitions, and tokens. Places are used to represent resources that are classified by functions. Transitions are used to represent the consumption of resources and the corresponding change of tokens. Tokens represent factors that affect system state, including raw materials, labor, equipment, data, and information. A set of tokens in specific places represents a state of the PN and is referred to as a marking. In addition, a PN may have an associated set of enabling and firing rules to determine under which conditions (i.e. a particular marking) a transition is enabled and may fire. In colored Petri nets (CPN) different kinds of tokens are distinguished by the addition of properties associated with a token. This allow for a more compact representation of real world situations. According to Qiao et al. (2002) a Petri net is valid if it has three properties: bounded, live, and reversible. Being bounded indicates the absence of overflow in the system model. This characteristic allows the specification of a limit on the number of tokens that may be in one place at any time. Being live implies that there is no possibility of deadlock. This characteristic is of significant importance in manufacturing systems with concurrent processes and shared resources, where deadlock conditions can easily occur. Being reversible indicates that the system can return to its initial state from any current state. This characteristic is very important for error recovery. The analysis of validity is difficult and time-consuming. Qiao et al. (2002) therefore propose a methodology for modeling valid CPN models. This approach defines some basic PN models whose validity is easy to prove. The manufacturing system to be modeled is decomposed into cells according to manufacturing function and the type of processing the cell will perform. Each of these cells is represented with an appropriate valid PN model. These models are sub-nets of the complete system model. According to a theorem of PN valid extension, the sub-nets can be connected using standard linkages or valid buffer models. As a result, the complete PN model of the manufacturing system is guaranteed to be valid. Models developed according to the methodology presented above do not solve the problem of how to validly add shared resources to the model. Shared resources are the main reason for deadlocks. However, two approaches that yield valid extensions with shared resources have been defined: parallel mutual exclusion (PME); and sequential mutual exclusion (SME). If shared resources are added according to these methods, the resulting model will remain bounded, live, and reversible (i.e. valid).

Falkman (2005) proposes a modeling language called process algebra Petri nets (PPN) to model resource allocation systems. PPN is a combination of process algebra and Petri nets. The high-level language combines the compactness of process algebra with the graphical features of Petri nets. In this way the language is capable of delivering concise and easy-to-read specifications for complex systems. That work focused on resource allocation systems, but the language can be used for other applications as well. Falkman (2005) furthermore developed a formal method for automated conversion of the high-level PPN specification to finite state automata. This eliminates risks for manual conversion errors at the same time that it enables the utilization of existing formal evaluation techniques for simulation, verification, and supervisory synthesis. Falkman (2005) also describes a method for enabling mapping between an information model according to the STEP standard (ISO10303-214) and a discrete event system specification using PPN.

4.6.6 Component variety, re-use, and commonality in manufacturing processes

Martin & Ishii (1997) propose a graphical representation that illustrates the flow of the product and its differentiation points (Figure 41). The figure illustrates the initial process sequence to the left and an improved sequencing to the right. Postponing differentiation until later in the process can help to reduce inventory costs and the complexity of the manufacturing system. This strategy is, however, not guaranteed to be beneficial in all cases. The potential benefits depend on many factors such as the lead times for procurement, manufacturing, distribution, and the customers’ desired lead-time. Martin & Ishii (1997) proposed that minimizing the number of nodes in the process sequence graph can be an appropriate measure. By minimizing the number of nodes, other measures such as inventory holding points and complexity of logistics are minimized.

Figure 41: Example of a process sequence graph (Martin & Ishii, 1997).

Martin & Ishii (1997) further propose a commonality index that determines the amount of commonality for each component:

For each process step, an index is calculated. An index value of one represents a common component that is used for all product variants. The component commonality indices are calculated for the left and right process sequences in Figure 41 and illustrated in Figure 42.

Martin & Ishii (1997) further discuss the lead-time required to purchase or manufacture a component. A component with a low commonality index and a short lead-time is less of a problem than a component with some variety (i.e. low commonality index) with a longer lead-time. A graph showing lead-time versus commonality index for the components provides an efficient illustration of these issues.

Figure 42: Component commonality indices (Martin & Ishii, 1997).

The case presented and discussed by Martin & Ishii (1997) provides an example of the speedometer with a rather low commonality index and a long lead time (see left graph in Figure 43). A solution to this problem by the case company was to split the speedometer assembly into two parts, the dial face (which the customer sees) and the motor and providing for manufacturing of the dial face at its own facility. In this case the company achieved a short lead-time for the variant part and a high degree of commonality for the remaining motor part with the long lead-time.

Figure 43: Lead-time versus commonality (Martin & Ishii, 1997).



Continue ...

Go to table of contents.