Infrastructure consists of facilities and services that are needed across an entire enterprise. Traditional infrastructure includes electric power, water and sewage, telephone service, and so on. For the next generation of manufacturing, information infrastructure must have the high degree of connectivity and compatibility that already characterizes traditional infrastructure. Key elements include database and information management systems, data communications networks and associated services, and applications software; the division into manufacturing-specific and nonmanufacturing-specific categories is particularly fuzzy for infrastructure elements. Information infrastructure presents a “weak-link” situation—the entire manufacturing system is only as good as the supporting information technology.
A vision for manufacturing in the year 2010 includes the ability to use an integrated information system to support the entire product cycle from beginning to end. Such a metasystem would facilitate communications among groups, automate “corporate memory,” and aid in making the various kinds of changes that occur routinely in manufacturing.
Paradoxically, although there is a requirement for tight integration among domains, there also is a need for isolation and tolerance of incomplete information to facilitate parallel development across domains. As an example, logic and technology design must be developed in parallel to some degree in order to meet time-to-market requirements. The logic designer must be able to use a cell from a gate-array library based on high-level parameters such as estimated area and speed even though the design of the cell has not been completed by the technologist. Moreover, data should be available, as generated, to anyone who needs them, regardless of where that person is.
Subject areas and research opportunities are listed in Table 4.1 and are discussed in the remainder of this chapter.
Architectures and Standards
As manufacturing equipment and systems come to share more common data and knowledge, the connecting interfaces become increasingly complex. Although each interface may itself be simple and robust, the myriad of interfaces in a large, diversified factory is fragile, subject to failure for relatively insignificant causes.1 As a result, applications developers spend disproportionate time and effort addressing the interfaces, per se, rather than the application itself.
TABLE 4.1 Infrastructure Systems
Example of Research Needed
Architectures and standards
Standard manufacturing control architectures
Generic functionality within control architectures
Cost-benefit criteria for potential standards
Data communications networks
Very high bandwidth technology and services
Target architectures and loads for shop floor networks
Transmission methods adapted to the manufacturing environment
Modeling and prototyping functions for user interface
Next-generation data manipulation language
Maintenance of data consistency and integrity through database updates
Architectures for autonomy and distributed intelligence
Autonomous agents to monitor and respond to production events
Knowledge agents for enterprise-wide management of models, names, transactions, rules, and so on
Architectures for manufacturing systems involving distributed intelligence
Tools to find and distribute information
Stable sets of rules for interacting agents
Dynamic variations in agent autonomy
Enterprise and inter-enterprise integration
Principles and architectures for coupling network-based applications
Automatic interpretation of transactions
Automatic message routing and associated processing
Support for multiple protocols and multiple speeds over a given medium
Time-critical message delivery in interconnected networks
Protocols and services that support specific demands of object-oriented applications
Services for human- and machine-based browsing and searching of information and resources
Tools and techniques supporting supply-chain dynamics and associated planning
Mechanisms and systems to support information session management
Example of Research Needed
Simplification of system designs, operation, and maintenance
Support for new programming paradigms
Tools for component-based architecture life-cycle approaches
General tools unconstrained by the limitations of specific programming languages
Tools with aspects of knowledge-based collaboration software
Rapid prototyping and other methodologies for faster development
Techniques for encapsulating legacy systems and developing mediator support
Analysis methodologies, metrics, and selection techniques
Reference architectures that cut across manufacturing domains to support software reusability
Data representations that depict both spatial and temporal aspects
Dependable computing systems
Better technology to support “hot swaps” of software
Continuous availability of online services
Fault-tolerant hardware and software
Increased system security and trustworthiness
Software, user interfaces, and hardware to support cooperative work
The vision for manufacturing in 2010 assumes that this problem has been overcome; connecting applications has become as simple as connecting household appliances to the power grid—one need only know how to run the application (like using a toaster) and use the interface (plug it in). This ease of interconnection and interoperation extends from devices found on the factory floor to applications connecting the factory to the product design facility to applications connecting an enterprise to its suppliers. An “application socket” for manufacturing would both benefit equipment vendors and enhance factory performance. Research is needed to make this vision a reality, including research relating to architecture, standards, and interfaces.
Open architectures can facilitate seamless connections from computer-aided design (CAD) to computer-aided process planning (CAPP) to computer-aided manufacturing (CAM). Architectures and standards bear on many levels of interconnection, from device-level to business-level. There are architectures for systems supporting enterprise functions, for enterprise-wide information flow, for control of processes, and for management. Complicating the problem of developing suitable architectures and standards is the need to combine commercial, off-the-shelf systems with layers of proprietary knowledge in the form of software and knowledge bases. The trend toward distributed intelligence, as well as the need
to form alliances with other business entities (including outside design and engineering services), also leads to significant knowledge management issues. Key questions include how to determine what proprietary knowledge to share through distributed intelligence and how to share it in ways that protect knowledge assets.
Research is needed to develop better architectures, including research to develop standard manufacturing control architectures (see “Equipment Controls” in Chapter 3) and generic functionality within the architectures, and to support general manufacturing information standards.
Standards are necessary to support the passing of information in the various architectures and the interconnection of different systems within the manufacturing enterprise. Standardization provides benefits of common systems (cost savings, improved integration (“plug and play” equipment and systems) and information flow, and so on).2 On the other hand, there continues to be a need to strike a balance between reducing the costs of variation and enabling rapid implementation of new technology—this is the fundamental tension over whether and when to standardize. Research could help to establish criteria for determining when the benefits of restricting options by a standard outweigh the costs.
The importance of standards for manufacturing interconnections raises the question of where research enters into the development of standards. The answer is not clear. For example, efforts to develop the product data exchange using STEP/standard for the exchange of product model data (PDES/STEP) are not research, per se, but the kind of problem that PDES is intended to solve may require research.3 Because of the importance of international standards in the global marketplace, strong participation of U.S. interests in standards making is valuable, and greater involvement of U.S. academic researchers (who are less involved than researchers in other countries) in standards making may strengthen the U.S. technology base in manufacturing. Research is a foundation for the consensus building inherent in standardization, and researchers should bring their insights to bear on that process.
Data Communications Networks
Communications will be the backbone of the future manufacturing enterprise. The ability to move large amounts of data quickly and cheaply will be equivalent to the need to move raw materials and products to manufacturing sites and distribution facilities. Infrastructure is needed to support that traffic, including a flexible computer network supporting a variety of applications, some of which have very high bandwidth requirements because of significant use of graphic images and the need for real-time control and communication.
Today’s factory and enterprise networks are not well suited to carry the profusion of data, images, and video resulting from widespread and rapidly growing personal computer use among engineers, management, and support personnel in the factory and adjunct support areas, as well as various types of test equipment with very high bandwidth requirements. At present, many manufacturing equipment interfaces are based only on RS232-D low-speed serial communications. Research is needed to support the development of very high bandwidth (hundreds of megabits per second to gigabits per second) networks supporting communication demands within facilities (as well as between them; see “Enterprise and Inter-enterprise Integration ” below).
Research is also needed to identify target architectures and loads for shop floor networks that will provide engineering performance references for design and performance benchmarking. For example, keeping the load down to 50 percent of capacity may be
cheaper and better for most applications than developing technology for managing or accommodating higher load levels.
Finally, research is needed on transmission methods, including wireless transmission, that are adapted to the manufacturing environment. Wireless communication is attractive to factories because it allows money, time, and effort to be saved when physical changes are made to the factory floor. Relevant research (and standardization) would address such issues as spread-spectrum-type wireless communications using trellis-type encodings to ensure maximum security, reliability, and noninterference with sensitive manufacturing equipment. Research also is needed to make more efficient use of the limited electromagnetic spectrum.
Modern database systems are designed to handle every aspect of data except generating or collecting them. These systems serve as the primary interface to data for the user through powerful query languages, for bulk storage devices through the file system, for the rest of the computer through the operating system, and for the rest of the world through communications networks. In each of these areas, advances have been rapid, but progress has only generated more ambitious goals. The various interface roles that a database management system serves provide a convenient taxonomy for the required research.
Overall, manufacturing applications are data- rather than computation-intensive.4 Enormous amounts of data are needed to support manufacturing activities (e.g., to represent the 1.5 million parts on a Boeing 777 or the millions of transistors on a VLSI chip), and the relationships among the data are complex, sophisticated, and changeable. For example, a VLSI component has a different description for each activity; functional, geometric, behavioral, temporal, and other views of the part must be supported. Any changes in one view must ripple through all of the other views and extend to invalidate or update usage of the component throughout a design. The movement toward multimedia databases and applications compounds the problems presented by simple text or numerical databases. The demands on database management systems for manufacturing are severe and will require a major research effort.
The demands that manufacturing makes on the user interface are particularly severe and present an opportunity for some creative research and development. The following two problems illustrate the type of work that is needed.
The first is the need to incorporate modeling and prototyping functions into the user interface. In particular, the basic syntax and semantics that describe products and processes will have to be incorporated in the data manipulation language of the database system. In short, the product and process description language should be a sublanguage of the data manipulation language.
Current data manipulation languages were developed to serve the financial community; they handle well financial and other types of data that can easily be cast into the relational model. Unfortunately, many manufacturing data types, especially geometric ones, are associative. Consequently, using current data manipulation languages for manufacturing data is counter-intuitive and difficult. To fill this gap, research is needed to develop the next generation of data manipulation language, perhaps based on the object-oriented data model.
A second major problem is how to maintain data consistency and integrity as a database undergoes constant updates. The problem is acute for both design and processing phases of manufacturing. During the design phase, the database must serve not only as the primary repository of ongoing work but also as the principal medium of communication among the participants (who may be using simulation, analysis, and design tools, as well as
manufacturing execution systems). Any disparity in the copies of the database seen by different users as changes are made will wreak havoc. The problem of consistency is also paramount in the processing phase, if real-time control of the process is to be achieved. Here the performance (i.e., speed) issue that is always present in database consistency becomes particularly difficult to resolve.
Architectures for Autonomy and Distributed Intelligence
Autonomous agents (implemented as software objects or collections of objects) are attractive for manufacturing applications in the areas of planning, monitoring, and control. One view is that autonomous agents, programmed by end users, could monitor production activities and respond to events with such actions as shutting down a machine, starting up a program, or sending a message to another agent.5 Because such agents automate control and interaction functions, they can eliminate activities otherwise performed by people and allow for simpler organizational structures, which in turn can simplify requirements for software development and maintenance. Another perspective refers to knowledge agents consisting of services and tools that could enable consistent, enterprise-wide management of models, names, objects, semantics, object relationships, object messaging, transactions, syntax translations, protocols, and business rules.6 The vision of manufacturing in 2010 might include an architecture for autonomous agents that represents each person, piece of equipment, and system in an enterprise by an agent.7
The promise of autonomous agents underscores the need for research to develop better architectures for manufacturing systems involving distributed intelligence generally. Distributed intelligence is fundamental to flexible, adaptive systems. Depending on the architecture, a centralized knowledge base and planner could make decisions and plans to be sent to resources within a manufacturing cell (reflecting conventional management hierarchies) or each resource (machine tool, vision system, tool, human being) could maintain its own intelligence about its capabilities and a global view of the environment. Research is needed to develop tools to find information (such as autonomous agents), tools to distribute information, and stable sets of rules for interacting agents (rules for agent behavior).
Research is also needed to develop agents that can act on an understanding of both global and local goals. The level of autonomy—complete autonomy, consensus (negotiation), or command-driven autonomy—is a critical consideration. Existing systems have focused on either command-driven or consensus-based autonomy. By contrast, dynamic variations of autonomy could provide a more adaptive system in terms of process, resource, product, and temporal requirements. Research is needed to enable dynamic variations in the level of autonomy.
Enterprise and Inter-enterprise Integration
Enterprise and inter-enterprise integration combines the methodologies and tools (e.g., communications networks) of electronic data interchange (EDI) with application software, data base management systems, and electronic mail systems. Within an enterprise, the goal has been to achieve linkages among design, production, marketing, finance, sales, human resources, distribution, and top management. A related goal is enterprise logistics management, from the pull of customer orders back through the entire manufacturing and supply chain. Although the vision of “computer-integrated manufacturing,” so potent in the early 1980s, has faded, enterprise integration, and within it the notion of what some have
called a virtual enterprise (dispersed components linked electronically, sometimes only for the duration of a specific project or program), has emerged as a broader goal. Implicit is the shift from a functional to a business process perspective.8 Progress toward this goal involves development of both physical (e.g., wires, gateways, and switches) and “soft” (e.g., protocols and information services) information infrastructure technologies.
Enterprise integration requires research and development relating to the interconnection of applications. Research is needed on organizing principles and architectures for coupling different network-based applications into a seamless environment. Such coupling is necessary, for example, to link flexible manufacturing cells to the plant scheduling function and to link the scheduling function to the enterprise order, delivery, and financial systems. Enterprise integration also implies a need for research to enable the automatic interpretation of the type of transaction being executed, the routing of the message to the right location for processing, and the processing that must occur when the message for the transaction reaches the correct system.
Current shop floor local-area networks include a dizzying array of communication technologies such as media, connectors, and protocols.9 Moreover, these local-area networks are connected to wide-area networks that include still more communication technologies. (Although there have been distinct differences between local-area and wide-area networks and their respective data, the demand for widespread concurrent engineering and general corporate infrastructure integration will require that many of these differences be resolved.) This proliferation of communication technologies not only prevents needed communication, but also raises maintenance costs. Research is needed to support a mix of protocols (e.g., transparent translation) and communication speeds over different media, or even (like the public telephone network) over the same media at a variety of speeds.
Research is needed to support time-critical communications systems, which are commonly used to control large-scale manufacturing processes. Such systems transmit control messages, each of which must be received and acted on in some critical time, ∆t. The magnitude of ∆t varies with the process being controlled, but the principle is the same as the Federal Express promise—the message positively has to be there on time. Time-critical communications are now generally transmitted on proprietary systems that assure timely delivery by restricting the type and number of messages carried. There is increasing demand for the ability to transmit control messages on existing open communications networks. But such networks use protocol suites such as TCP/IP or Open Systems Interconnection (OSI) that deliver messages on a “best-effort” basis and cannot reliably meet delivery-time requirements. Thus research is needed to formulate the principles for construction and operation of networks that support time-critical message delivery in a context of interconnecting, multipurpose networks. No one knows how to do this because the requirements for such a system are not well characterized.10
Research is needed to develop protocols and services that support more efficient object-oriented intercommunication, remote object invocation, interobject class and metaclass exchanges, lightweight message passing, and other specific demands of object-oriented applications that are not well accommodated by today’s typical local- and wide-area networks. In addition, research relating to data compression will continue to be important.
Research is needed to develop services tailored for human- and machine-based information/resource browsing and searching, using knowledge-based assistance agents for semantic interpretations, translations, and relationships. These essential services should become part of the underlying network service infrastructure in order to increase network performance and efficiency. The higher-level network directory services of today will not meet the requirements of near-future applications.11
Increasingly, manufacturing organizations will be coupled to their suppliers, customers, and other partners via electronic threads;12 this is the essence of inter-enterprise integration. At present, much of material-related production cost stems from the packaging, shipping, unpacking, and moving around of component materials. Better communication between supplier and customer enterprises would reduce such costs; although not new, EDI and other tools to support “just-in-time” or other streamlined processes are obvious points of departure and candidates for supporting technology. For example, a customer’s highly customized order for even a very complicated and sophisticated part (for which there may be many alternative choices) could be entered directly by the customer into a manufacturer’s system, the customer order would be scheduled, and the customer would receive immediate acknowledgment with a product delivery date.13 Similarly, a request for material replenishment could go directly and electronically from a manufacturer’s factory floor to the supplier’s system. The supplier would deliver the material directly to the factory floor, where receiving activities are performed. Both of these transactions could be associated with the electronic communication of financial information and even electronic funds transfers, involving corporate-level systems and systems at third-party financial institutions. The interconnection of supplier and customer enterprises implies a need for research to support supply-chain dynamics, including data acquisition through distributed systems. In addition, tools and techniques are needed to enable manufacturers to plan on a real-time basis back through the entire supply chain. This level of analysis is generally impossible today; given the typical operation of manufacturing resource planning systems, it can take weeks.
Mechanisms and systems are required to support information session management. These include the means for connecting to and coordinating the delivery of information between multiple sources using multiple streams, as well as the mechanisms for collaboration with this information. Research will be required in terms of software architectures, protocols for communication, and access and security mechanisms (see “Dependable Computing Systems ” below). These needs are above and beyond the transaction-based processing and management that will clearly be required.
In view of the experiences of the major manufacturers represented by members of the committee and those who briefed the committee, the current state of the art in software engineering is barely adequate to meet current manufacturing needs. Several companies have invested many thousands of person-hours in developing and integrating factory information systems, a disproportionately large investment in comparison with their investment in other parts of the factory system. This imbalance is especially serious considering that these companies have often purchased “finished” software application systems from commercial sources, which when measured against the benefits perceived have proved to be less than satisfying. Large-scale systems for manufacturing present special problems, since they are often one of a kind, are developed by teams with relatively limited experience in exactly the kind of system needed, and generally entail very high life-cycle costs.14 Commercial software does not integrate well with software from other commercial sources or with internally developed software; there is little or no access to source code to make minor but important changes; revisions come out slowly and hence frustrate users who expect rapid response; and so on. Customizing of general applications to specific customer needs and
integrating applications and technologies from a variety of sources make development of a robust manufacturing system a formidable task.
In the 21st century vision of manufacturing, trained manufacturing personnel, rather than software development experts, would be able to develop and change application systems for the shop floor or the design laboratory. This capability means that both better methods of developing software and better human-machine interfaces are required to enable domain-specific software specification.
Research is needed to support faster development of easier-to-use and more effective systems. Simplification of designs, operation, and maintenance is desired, as is increased predictability of systems, self-healing systems, and system extensibility. Research is needed not only into reusable components (see “Reusability of Software Design and Analysis Artifacts” below) and system analysis and design (see “Analysis” below) but also into system optimization and enhanced system operation and maintenance. Visualization and human-computer interaction techniques will be key.
The need to accommodate new programming paradigms also should drive research. For example, the current emphasis on object-oriented programming (see Appendix A) suggests a variety of research topics; there also should be consideration of inevitable new approaches to programming.
General Tools for Software Engineering
Automating complex, intelligent manufacturing systems requires both engineering life-cycle approaches and supporting tools. Tools are needed for system analysis, requirement management, design, configuration and integration, and simulation.
Software engineering tool development lags significantly behind the use of new methodologies in many cases, raising the cost of building large, complex manufacturing systems. For example, better tools are needed to support component-based architecture life-cycle approaches. Also needed are more general tools that are not constrained by the limitations of specific programming languages, as well as tools that have aspects of knowledge-based collaboration software.
Because changes to products and processes occur frequently, manufacturing software is subjected to constant stress. Research is needed to support faster development of systems (both to meet ongoing needs and to facilitate the transition to newer technology), using rapid prototyping and other methodologies. The need to accommodate legacy systems can be addressed by research to develop techniques for encapsulation of legacy systems and to develop mediator support.
Although there are many different analysis methodologies, such as responsibilitydriven, data-driven, and activity-based methodologies, system analysis remains more an art than a science. How do we compare types of analyses? Are some techniques simply better than others, or do different types of system development or applications dictate a choice?
Are the techniques sufficient, or are new ones needed? Research is needed to help answer these questions.
Within the area of analysis, research is needed to improve metrics. Metrics typically are developed for the testing phase of the engineering process to determine system performance in terms of speed, timing, and so on. There are no metrics that specify how well the problem is understood and formulated. What are the metrics for analysis, and how do we measure our successes and failures? When do we know that analysis is complete or at least good enough? Can these questions be addressed with metrics?
Given the frequent need for rapid system development, metrics specifying temporal aspects of analysis should be considered. Are there temporal metrics for analysis techniques (such as state transition diagrams, object models, and event-stimulus diagrams) that could be related to system characteristics?
Reusability of Software Design and Analysis Artifacts
The reuse of software and system artifacts and results from associated analyses in the development of new systems or applications would leverage previous investments in expertise, effort, money, and time. Examples of artifacts from object-oriented analysis and design include object models (including representations of data and behavior) and object interface definitions. Reuse of software system elements has been an elusive goal; research in this area would benefit manufacturing as well as other application domains.
One type of research supporting reusability could lead to reference architectures that cut across manufacturing domains. For example, there could be families of reference architectures for continuous processes and for discrete processes. Goals would include minimizing the number of reference architectures for which third-party suppliers must develop software or systems and increasing the ability to leverage previous expertise across projects.
Increased Spatial and Temporal Dimensions of Software Representations
Software representation currently focuses on declarative knowledge. Research is needed to achieve representations that depict both spatial and temporal aspects of associated data, enriching a system’s repository of knowledge and facilitating the visualization, design, specification, monitoring, and analysis of manufacturing processes. Multimedia visualization of these new representations may include “walking through” models or providing simulations of system behaviors in one window while the activated models employed are presented in another window. Such elements would contribute to the realization of virtual factories, which involve enhanced modeling and simulation of processes and products. New programming interfaces that capture the spatial and physical abstractions of manufacturing are essential to allow end users to program their own processes and workflows.
Dependable Computing Systems
Manufacturing plants require continuous operation, creating a need for dependable computing systems. Better technology to support “hot swaps” of software (i.e., changing software without removing the system from operation), continuous availability of online
services, and fault-tolerant hardware and software are among the technologies needed for dependable manufacturing systems.
Manufacturing provides an application arena for a wide variety of research relating to increased system security and trustworthiness, including access control and authentication. The need to determine and verify the user of a computer system or network is becoming increasingly important as a way of preserving system and data integrity, ensuring that sensitive data and systems are accessed only by those who are authorized, and ensuring the highest levels of system availability. These concerns affect both intra-enterprise and inter-enterprise communications.15 The manufacturing environment, especially the factory environment, calls for economical and robust technology to address these needs.
Collaboration Technology/Computer-Supported Cooperative Work
The trend toward organizing workers of all kinds into teams with significant levels of decision-making authority gives rise to a need for technology to support collaborative activity.16 For example, intelligent systems are needed to support collaborative efforts in the design of complex products; they can also facilitate collaboration among factory and other, nonproduction, personnel. Research needs include information technology to support empowered work teams of various kinds of personnel and tools for total quality management.
Research relating to technical tools for computer-supported cooperative work (including software, user interfaces, and supporting hardware) should be complemented by research examining relevant aspects of human behavior, education and training requirements, and so on, to ensure both that optimal tools are developed and that they can be used easily.