Enabling Technologies for Unified Life-Cycle Engineering of Structural Components (1991)

From its study of today's environment covering the extent of unified life-cycle engineering (ULCE) technology and application for structural components—Coupled with predictions of the future environment and potential for ULCE, the committee identified a list of issues raised by comparison of current and future environments. The critical issues are:

• ULCE-driven development of materials processing and repair methodologies requires integration of research and development (R&D) across disciplines.

• Advanced analytical modeling and simulation methods to predict actual component manufacture, operation, and logistics do not exist to the extent required to preclude the need for physical prototypes and mock-ups.

• The information system for an integrated team approach to ULCE is inadequate.

• The ULCE team will need to make key decisions while still operating with incomplete information.

These issues were then distilled and used to develop a set of needs and concerns. The needs and concerns were reviewed to extract the underlying critical issues and are the basis for the committee's recommendations.

The related needs and concerns and their associated critical issue are shown in Figure 7-1. From the needs and concerns a set of enabling technologies needed to address each issue was developed and the enabling technologies and critical issues they support are shown in Figure 7-2.

Two case studies were undertaken to evaluate the concerns, one on a metallic turbine disk (Appendix A); the other on a fiber-reinforced composite horizontal stabilizer (Appendix B). In

the case studies it was found that all of the needs and concerns identified by design engineers were associated with at least one of the four critical issues defined by the committee. The comparisons support the validity and completeness of the generic critical issues developed by the committee. Additional insight into the benefits and requirements of the ULCE approach is obtained by understanding the needs and concerns which are the basis for the critical issues and a brief description of each and of the enabling technologies is given in the following section.

This issue highlights the integral role of materials science, development, processing, characterization, and engineering in ULCE. Most important is the recognition that knowledge of materials processing and durability strongly affects design decisions. Knowledge of the manufacturing processes and the resulting materials properties is required at design time. Coordination of materials research and development with ULCE design and manufacturing needs is viewed as critical to ULCE implementation.

The science and development of new materials is progressing rapidly, resulting in a reservoir of materials with potential for significant performance improvement. It is inevitable that many of these materials will be pressed into service before their performance is fully characterized. Complete characterization is extremely difficult because of three factors: incomplete specification of the environment in which the material is to perform; lack of fundamental understanding of materials behavior; and the long testing time necessary to conduct the evaluations.

Even during the initial stages, material under development is targeted for a particular component family of applications (e.g., turbine rotor disk forgings or turbine nozzle flap skins). This often occurs before a specific weapons system application is defined. ULCE needs for component producibility and supportability in the field can, and should, be addressed at this earliest material development stage.

Key technological issues that must be addressed prior to application of materials are discussed in the following sections. However, the overall strategy needed for ULCE to capitalize on gains in materials development is the coordination of materials efforts with design and manufacturing early in the product development cycle.

The lack of information about product and material variability as processed is a serious concern. To predict the behavior of a material in use, it is necessary that a broad range of material attributes be known; this sometimes requires an exhaustive set of manufacturing data in addition to laboratory data. For many "conventional" materials, information on behavior may be incomplete or difficult to obtain for different application configurations, particularly in sections of very large or small sizes where the alloy microstructure and phase distribution are different from those typically obtained, or for sections (like weldments) having gradients in microstructure.

Thus, structure-property correlations need to be extended to include the effects of processing variations, particularly for components with extraordinary materials microstructure. This information needs to be included in materials data bases. For new materials, particularly ceramics and composites that are very sensitive to defects that may be introduced by manufacturing and processing, the information available is usually sparse and inadequate for design use.

The effect of the service environment including temperature, stress, corrosion, embrittling conditions, battle and accidental damage on materials behavior needs to be formulated in a framework usable for ULCE. Understanding of the behavior of conventional metal alloys exposed to simple test conditions (e.g., uniform temperature and stress states giving steady-state deformation) has increased rapidly over the past 3 decades. This knowledge base has allowed formulation of constitutive equations describing observed behavior, and sometimes predicting it. Actual service conditions are far more complicated than these test conditions, particularly when time-varying conditions occur under which the principal modes of damage change over time as well. The dependence of damage by one mode on previous (or subsequent) damage by another mode is not understood in conventional alloys, much less in emerging ceramics and composites. In addition, materials behavior under transient conditions is poorly understood. Of special importance is the transient leading to failure, which is usually controlled by localized nonuniform events. Understanding of the physical nature of damage and its quantitative formulation are needed. While efforts must be devoted to both model development and experimental programs, emphasis should be placed on modeling because of the huge data base needed if it is obtained through empirical testing. For example, creep deformation involves several independent variables such as temperature, stress, frequency, and amplitudes of transients. The modeling recommended would aim toward formulating constitutive equations that combine first-principle results with engineering considerations. The experimental efforts considered necessary are those characterizing the physical nature of damage (e.g., arrays of dislocations, cavities, grain boundaries, and other interfaces). It is necessary to describe not only individual defects but also the statistical or probabilistic nature of their distribution and ultimate influence on properties.

The ability to extrapolate materials performance is imperative, since it will not be possible to develop a data base for all contingencies. The accuracy in extrapolation will depend on successful characterization of the manufacturing process, testing of the as-produced material, analysis of damage accumulated in service, and correct specification of future service conditions.

Extrapolation or life-prediction models will need to extend damage models to include a criterion for incipient failure and termination of service. This extension is nontrivial because, whereas damage occurs initially in a quasi-homogeneous manner throughout the material, failure is normally controlled by a series of localized events that usually occur on a statistical rather than deterministic basis. Therefore, life-prediction models in most cases need to be probabilistic. In addition, as design requirements approach the incipient failure point of the material, nondestructive evaluation (NDE) becomes increasingly important to detect the precursor state of the "critical" defect. This topic is discussed further under the second critical issue on modeling methods.

Experimental data are needed for various materials science and engineering purposes to assist in understanding fundamental materials behavior and in validating models thereof; to guide the development of advanced engineering alloys and novel materials; and to establish valid data bases for design, manufacture, and product support. A hierarchy of test methods will be useful, with the complexity, cost, and standardization usually being greater for engineering tests than for science tests.

Closer control of materials processing is required to obtain reproducibly and reliably the desired material microstructure, component structure, and properties. One approach to achieving this control is termed intelligent materials processing, where sensors, process models, expert systems, and closed-loop controls are brought to bear on specific processing operations. This approach needs to be developed for materials processing in both component manufacture and repair. Improved processing control will also hasten commercial availability of novel or new materials, lower the variability of as-produced properties, and improve the yield.

Given that several different processing sequences can be used to produce a component of a given configuration and that a large number of configurations is sought for various applications, it would be impossible to examine and record data for all possible combinations. Therefore, a new approach is needed in applied research and engineering to express results on the basis of materials phenomena (e.g., thermodynamic or kinetic terms rather than process-specific terms) so that results from one study can be transferred to others. This basis has already been adopted for some aspects of materials processing, such as modeling solidification and deformation.

Modeling the overall materials manufacturing sequence is a critical need in the development of ULCE. This modeling would need to extend efforts on materials processing to include the reliability of multiple-step processes as well as other aspects such as inspectability. These manufacturing processes must address both initial component manufacturing and in-service repair, since field military equipment will be subject to environmental damage, battle damage, and maintenance damage, as well as normal wear and tear. It is imperative that appropriate repair techniques be developed that can restore most of the original structural capability. Where damage is extensive and/or severe, these repair processes may be more appropriately described as remanufacturing processes. In combat, these repairs must be carried out under significantly less than ideal conditions.

Central to the critical issue of ULCE-driven development is the fact that uncertainty in knowledge of materials behavior is becoming more costly as demands for greater system performance increase. To respond to these needs and concerns in an ULCE framework, the following enabling technologies are essential. In developing these technologies, materials efforts must be integrated with design and manufacture.

Three aspects need to be emphasized--sensors, modeling, and correspondence between field exposures and controlled test conditions. Sensor capabilities have improved dramatically in the past decade in terms of durability of the sensor, reduced size, and sensitivity to various phenomena. However, sensor development has been oriented toward commercial markets such as automobiles, medical devices, and buildings and their usefulness in military structural components needs to be evaluated. In addition, the development of advanced sensors to detect microstructural changes in materials via NDE (e.g., fiber-optic sensors embedded in fiber-reinforced composites) has promise.

Progress in modeling materials behavior has been more successful when behavior is controlled by a single phenomenon or defect, whereas models of behavior dependent on cooperative phenomena have not been able to predict material property values and sometimes not even trends in properties. The major obstacle is the lack of physical understanding of the interdependence of the phenomena involved and continued attention needs to be devoted to this difficult problem. The availability of large computers will be helpful to facilitate sensitivity analyses of the relative influence of different variables on others. Without valid models, much of the output from sensors will not be utilized effectively.

Establishing correspondence between field action (e.g., maintenance or battle damage) and controlled tests is a prerequisite to valid prediction of field damage in a form that can be used quantitatively in design. Greater effectiveness is needed in two areas: tests and simulation of components to model performance and application of sensors to the structure or components to monitor performance. This, coupled with simultaneous measures of service conditions would provide a basis for model validation and provide input data for future analyses.

Advances in repair methodologies, inspectability, and lifetime prediction models are needed to enhance design for supportability. Also needed are the advances described for damage characterization. Lifetime prediction models that are needed for supportability to ensure timeliness of maintenance and replacement are treated more thoroughly under the next critical issue.

Emerging technologies that will improve component inspectability include sensors, NDE, and three-dimensional computer imaging and analysis. Embedded sensors will allow more thoròugh documentation of actual service conditions encountered. Sensors monitoring material structural integrity will promote more effective NDE of components, while computer imaging and analysis of components during design could identify optimal locations for sensors and indicate the suitability of different regions for interrogation by various NDE probes. The overall goal of these efforts should be the development of guidelines (although qualitative) for design of generic classes of components using state-of-the-art sensors and NDE.

Limitations and capabilities of repair methodologies need to be considered during design since, clearly, more flexibility and control of repair processes will promote supportability. A promising direction toward these goals involves automated materials processing using a closed-loop approach that combines in-process sensors, computerized data bases, and control systems. Inspection during processing is relatively new and is intended to help avoid the high costs of rejecting fully processed material. Although this approach has been proposed mainly for initial processing of materials, the concept is equally valid, albeit more difficult, for repair methods. An automated materials processing facility may be visualized as consisting of four principal

interconnected systems. The first system is the processor itself, such as weld overlay apparatus for repair of metal alloys or a mold and autoclave for repair of polymer matrix composites. The second consists of sensors that can measure or monitor important properties or characteristics of the material during the processing step. The third, of control algorithms that regulate various experimental parameters. The fourth is a computerized data base that provides information that potentially can close the loop between the sensors and the process controls and replace the human component.

Advanced analytical modeling and simulation methods to predict actual component manufacture, operation, and logistics do not exist to the extent required to preclude the need for physical prototypes and mock-ups.

ULCE must include the ability to predict the economic consequences of change. Decision-makers need to be able to understand how product design change, modification to a process (manufacturing or support), or an extension to an operational mission affects the life cycle of the product. The impact is measured, to a large extent, in cost. Therefore, an economic representation of the life cycle of a product is a cornerstone for all other modeling and decision-making efforts. In addition, there are also educational and training benefits from the development and use of ULCE cost models.

The life-cycle cost model should be available before conceptual design begins--at the time when product requirements are set. The model should be able to evolve through the life cycle along with the product and be refined as the product, manufacturing, and support definitions are developed. An ideal cost model should be robust enough to allow new technologies, methods, and rules; stable enough to allow back-tracking of decisions; and flexible enough to allow details to be added as they become available. The model needs to include all components of life-cycle costs (development, manufacturing, and support) and should conform to the accounting practices used for decision-making.

In addition to the economic view, designers need to compare their product design to the functional specifications in a way that accounts for variation in materials, manufacturing, operation, and support. The following particular capabilities can improve lifetime component performance:

• Residual life analysis for components and assemblies to test for malfunctions during operation that are not observable during manufacture. These models require an increased gathering of knowledge about use conditions and prior failure mode, with wide dissemination of the new information.

• Simulations of stochastic materials behavior, taking into account local behavior and failure initiation to provide more accurate predictions of material performance in manufacturing,

support, and operation. These models are needed to predict the manufacturing and support costs and to reduce the product development cycle time.

• Models to predict the damage from operation and various threats for each class of structural components. These models are needed so that complete life-cycle stress analysis can be computed. Stochastic methodologies to describe damage threats are also required.

• Integration of statistical process control models with process models to assess process capabilities at design time. These models are needed to provide a basis for ensuring the stability of the manufacturing process implementation, especially for adaptive control applications.

Validation of the applicability of mathematical modeling and simulations to a given component service environment is required. Before liability questions can be trusted to automated methods, a sound methodology is needed for determining and certifying the accuracy and limitation of models and computer simulations.

To obtain good manufacturing feedback during design and, in particular, during conceptual design, more capable and comprehensive modeling and analysis of manufacturing processes are required. Software that can simulate a manufacturing process (e.g., forging, forming, casting, machining) not only provides an opportunity for reduction in manufacturing costs associated with tooling and trial-by-error process development but also can provide designers with process choices early in design. In the past 5 years, computer-aided engineering (CAE) tools have come into use for manufacturing applications. The potential for manufacturing CAE is very large but needs further development. Since current methods of CAE were developed for product design and extensions are needed to account for unique manufacturing needs. For example, large material strain induced by deformations in manufacturing require special attention in finite element modeling and solution algorithms and boundary conditions for processes are, at times, more complex than those for modeling a component in operation.

Higher-level simulations of factories and sequences of manufacturing operations offer considerable promise in providing a product designer with trade-offs and impacts of a potential design. Today, these simulations are used to develop manufacturing strategies and capital plans. These activities are not generally part of product design but are pursued after design. In the ULCE environment, simulation of manufacturing operations is needed as part of the initial design effort; however, extension of today's factory simulation capability is required for ULCE.

Manufacturing CAE and manufacturing simulation have been demonstrated and are in use to some degree. The extension of these techniques to both support and logistics applications also has great promise. Since the use of these techniques is not widespread, the potential of this application area should be identified for CAE and facilities simulation suppliers. For example, designers in an ULCE environment need to assess the impact of their designs on the necessary support facilities since simulations of repair facilities can give a designer inputs and trade-offs that can be factored directly into a new design.

The introduction of the ULCE approach into the design process will have significant impact on the role of today's designers. They will be asked to do more than today while still

maintaining design costs and schedules to near-current levels. Careful analysis of the present design process is required to effectively achieve ULCE. Numerous productivity efforts (e.g., IPAD and CAD-CAM) have been initiated to address design efficiency, yet the demands on a designer remain very high. The use of the computer as a design tool can be expected to increase. Thus the roles of designers, senior designers, managers, and technicians need to be reviewed and recommendations on the best design organizational structures in which to implement ULCE need to be made.

The technologies that are available, that can be developed, or that need to be started to address the critical issue of modeling and simulation are described in the following paragraphs.

An economic analysis of the life cycle of a design can be satisfied using existing technologies and methods. There needs to be a coordinated effort among government agencies and contractors so that algorithms for life-cycle-cost components are standardized and rates and factors for distribution and support costs can be provided in a common framework with common understanding. (Examples are labor rates for repair centers, shipping costs for spare parts, and power-on hours for machinery.) Such an effort would include a methodology for calculating the life-cycle cost for components, subassemblies, assemblies, and products. The resulting algorithms and definitions could be made public so that software vendors could develop commercial-quality software for use by government contractors.

Another important aspect of modeling and simulation as a critical issue is the need to be able to predict the lifetime performance of a design. Lifetime performance includes manufacturability, operability, durability, and repairability. Comprehensive performance prediction is extremely complex and requires the application of numerous sophisticated technologies, some of which are reviewed in this study. One way to ensure the performance of a design that builds on past applications is to use proven features and approaches where possible.

To ensure the integrity of new structures, there will continue to be a requirement to analyze the design for stress-strain relations in response to external loads. Finite element methods have matured and are in common use but are only available for deterministic analysis. Thus, variations in materials properties, dimensions, and applied forces can only be considered by repeating analysis computations with different parameter values. This approach is expensive and leads to output that is voluminous and difficult to confirm experimentally or to use to estimate the stochastic behavior of the design. However, stochastic finite element methods have been research topics for the past few years, and the technology is ready for development and deployment in a prototype computational system.

Although this finite element analysis capability can provide an analysis of deformation for specific applied forces, the magnitude of the applied forces needs to be available as an input to the finite element model. This is particularly difficult for dynamic systems because they usually operate across multiple energy domains. An example of a promising technology that provides a unified view of system dynamics across all of these energy domains is bond graphs. Currently,

computational software for bond graphs is limited to certain classes of problems because of limitations in numerical mathematics and automation of bond graph-to-equation algorithms.

A cornerstone of a computerized design capability is the ability to develop complete product definition descriptions and utilize these descriptions throughout the product life cycle. This capability needs to support the evolutionary buildup of a design from early (conceptual) design stages through manufacture and support. Today's CAD-CAM systems do not fully meet this requirement. Deficiencies lie in the inability to represent all the product definition data since CAD-CAM systems, including second-generation solid modelers and workstation CAD-CAM systems, focus primarily on the geometric data and are deficient in handling nonshape data (i.e., notes, tolerances, specifications).

A key aspect of the modeler is its ability to relate process information to design definition. A desired system would be a combination product and process modeler that contains information on producibility tightly related to design attributes. In this way, designers can see the manufacturing impacts and constraints of their design options. The process data would be provided from manufacturing experience. For example, turbine disk standard hole sizes and finishing processes would be available to the designer in order to specify the type of hole in the new design. With this technique, the best designs and manufacturing efficiency can be enhanced and maintained.

A component of the product-process modeler is feature-based modeling, which enables designers to use proved ideas to ensure the manufacturability and performance of the item or assembly. Feature-based modeling also has inherent advantages over conventional dimension-based design:

• The manufacturability of the design is assured, and the development of a process plan can be automated.

• The specification of the design is less ambiguous and more complete than with dimension-based design. The designer is prompted only for the required parameters for the feature and for no spurious or conflicting information, as often happens with conventional dimension-based design processes.

Feature-based modeling is an emerging technology. Some feature-based systems are built on three-dimensional solid modeling systems, and others are based on object-oriented systems. Current research on feature-based modeling has focused on manufacturability in the definition of features and parallel effort to develop durable, reliable, repairable features should be initiated. A preliminary study is necessary to develop at least a first list of ''good'' features from a support perspective, while refinement of the list could proceed in parallel with research to utilize the support-oriented features during design.

The information system for an integrated team approach to ULCE is inadequate.

The needs and concerns resulting in this critical issue fall into several categories. The first is concerned with the ULCE system structure and functions; the second is related to more specific software requirements for information system components. A third category is concerned with organizational aspects. A fourth is the limited power and memory of mainframes to be used in the solid modeling process of design. Supercomputer systems need to be adapted to support solid modeling as utilized in the design process to support the definition of data for manufacturing. A primary concern is that the needs of the ULCE information system cannot be met adequately merely by writing more software. Indeed, more software packages may be counterproductive if the desired system-level reference models and architecture are not present.

Generic needs and concerns common to all software systems and information systems are not discussed here. Only ULCE-specific matters are addressed. The others may be important but undoubtedly will be addressed as technology moves forward on a broad front.

The primary need and concern for ULCE information systems hinge on the agreement of all involved parties on how the ULCE system might, or should, work. At the system level, there is need for an information reference model of what a working ULCE system might look like. The model would provide a description of the functions and information needed in a prototype ULCE system and would help establish a common basis for learning and commenting on ULCE. The model would illustrate how individual functioning components of present systems could be changed and would show the interrelationships between the components. This effort is needed to carefully comb out and describe the procedures carried out, the considerations governing these procedures, the strategies, the precautions, and the checks. The concern is that, in the absence of definitive models of ULCE information requirements, perceived deficiencies of the system would be met with costly attempts to provide misguided patchwork solutions, which, although powerful in some local sense, would merely further confuse the issue and impede real system-wide progress.

Although an English-language description of the ULCE system might be very valuable as a first step, it would be of somewhat limited utility. There is a need to "tell it to the computer"—that is, to formulate a description or, essentially, a simulation of the system in software. One can then exercise the dynamics of the model to explore the consequences of this understanding. Because business and technical requirements change, the reference model and the ULCE software systems need to be easy to change, tailor, and evolve to fit particular circumstances.

Because of the complexity and scope of the software systems that describe ULCE systems and support ULCE systems tasks, it is essential that the conceptual structure of such systems mimic as closely as possible and to the extent useful, the thought processes and work habits of the humans who use the system. In other words, such software systems should be "intuitively" understandable to their human users insofar as interface interactions are concerned and insofar as thought processes and expectations are concerned. This lack of "intuitively" understandable software has hampered full exploitation of today's software and will be critical to ULCE system success.

To the extent that such understanding or "user-friendliness" can be implemented, there is hope that information management for task support in ULCE can be implemented in an effective way. The system would be able to edit and update information bases in a meaningful manner, grant access to information in appropriate ways, and remind users of related information and checklists at the proper junctures.

Design assessment tools need to be available from conceptual design through field upgrades. Designers need to be able to determine whether the manufactured system will function in service and if so whether it will perform reliably.

Software needs include design assessment tools incorporating powerful finite element packages for structural, thermal, and aerodynamic analyses; simulation tools; and capabilities to model and display components to aid in visualizing designs. Similarly, there needs to be software that will allow manufacturing process information and software to be meshed more efficiently and flexibly with information and software from design, manufacturing facilities, materials, and field experience.

In the design process, aids are needed to track the process, set schedules and procedures, provide design information, verify manufacturing assumptions, and provide general information on preferred design strategies. An additional design need relates to materials technology—to track experiences with materials and associated fabrication technologies. Design of materials, treating procedures, and data discrimination will be included in this mode. Maintenance simulation models have the potential capability of simulating many maintenance actions. These simulations would eliminate the need for many physical prototypes.

Fundamental to ULCE is the development of a teamwork culture and methods of management that closely integrate design, manufacturing, and support. Software development alone cannot provide the ideal ULCE environment if data and decisions are not transmittable across technical and managerial boundaries.

Training and education of the ULCE team in system operational philosophy, strategy, style, and accountabilities is also a concern that should be addressed. In particular, it will be necessary to cross-train at least key members of the ULCE team to learn about operating and maintenance conditions. To be effective, the ULCE data base and design rules will contain information relating to reliability, supportability, and producibility; without proper training, these would be of limited use.

Formal education providing perspectives on each of the functions of design, manufacture, product support, materials, and information flow, as well as on the integration of these functions in the ULCE environment, is needed.

The development of a powerful, integrated computer environment for ULCE could accomplish much, but its impact will be diminished and could even be negated unless organizations interact in an appropriate manner. While it is difficult to anticipate every situation several examples serve to illustrate the point. Simultaneous access to the evolving design by all departments could lead to chaos unless rules governing design modifications are in place.

However, the simplest rule, sequential reviews after design completion simply recreates the current paper process in electronic form. Likewise if government auditing tools for design review and approval required a paper-based review and approval system, the difficulties or fragmented data bases that it is designed to resolve would reoccur.

The new disciplines of artificial intelligence and neural nets and their specialty subareas based on symbolic processing, nonprocedural programming languages, and use of heuristics need to be motivated to fuse their strengths in order to handle practical problems of significant size and technical difficulty. ULCE is one such endeavor. For example, the science and technology for modeling large complex systems and for implementing a computer simulation of that modeled system need to be developed for ULCE. Such a model and simulation would be quite different from those that would be obtained from existing systems. The model needs to be modularly represented and intuitively understandable, be capable of representing and accommodating many diverse types of knowledge sources, be capable of representing the dynamics and temporal evolution of such systems, and have effective interfaces.

The performance of computer software systems will not totally meet human expectations unless there is knowledge in these systems of what humans would naturally expect in the way of reasonably intelligent competent behavior under various circumstances. So-called conceptual dependency theory and related investigations have made some progress toward implementing understanding in systems and building memories in systems in such a manner that appropriate associated events can be recalled on cue (Sowa, 1984). Development of such technologies is needed for implementing "understanding" and "learning" capabilities in systems.

Machine learning technology, of which there are several varieties, offers promise. In deductive learning, the intelligent system carries out deductive inferences to establish explicitly facts or rules that are implied by existing data. This capability is important because, in almost all cases, a fact or a rule needs to be made explicit before it can be utilized fully. Deductive learning is the best-understood and best-developed of all the machine learning technologies. Inductive learning is not only difficult but also risky, because the next example may prove previous inductively inferred knowledge to be wrong. However, machine learning can also be achieved with use of connectionist nets or neural nets. This technology has its limitations in that all input needs to be represented in terms of numeric values. For example, a neural net can observe what actions should be taken for a number of specific circumstances and can then generalize on that information to specify the actions that would be appropriate for circumstances that had not been considered before. If it is wrong and is informed of that fact, it would correct its own knowledge to yield improved responses in the future.

Present-day data-base technology is robust but still is not flexible enough. Technology needs to be developed so that data bases can deal with a wider variety of data formats and data types (e.g., text, voice, cursive, video, numeric, graphic, and telemetry). In addition, given "understanding" and "learning" capabilities, data-base management systems could store and retrieve information in ways that allow all expected associated actions to be taken and permit

communications to take place between different data bases supported by different systems. Semantic data-base research seems to be progressing, but not rapidly enough.

The current industry standard for product definition data exchange is the Initial Graphics Exchange Specifications (IGES) developed by a volunteer industry committee supported by the National Institute of Standards and Technology. IGES acts as a common format for product definition, transferring geometric images from one computer system to another.

Although product definition standards have been evolving with IGES and Part Description Exchange Specification (PDES), technology needs to be developed to describe products in feature-based formats, using three-dimensional structures. These efforts will extend the current IGES and PDES formats to incorporate actual manufacturability and supportability features, forms, tolerances, surface quality, and assemblage. Developments in the extraction and representation of manufacturing features can have a significant impact on product development and manufacture.

The ULCE systems considered in this study are distributed systems, and effective communications are essential for their support. The Manufacturing Automation Protocol (MAP) developed by General Motors and the Technical and Office Protocols (TOP) developed by Boeing are promising communication efforts that should be followed and encouraged. MAP, designed for the factory, and TOP, designed for the office and design environments, are being developed to be mutually supportive. These protocols help establish the computing framework by which all the ULCE needs are addressed. Specializations of ULCE might be necessary and should be considered. In particular, extensions to MAP and TOP, which support logistic and field applications, may be necessary.

The collection and reduction of data from the field produce only a limited sample of possible failures. However, the most important but least reliable source of information about actual reliability and supportability is obtained from field feedback. Field operations result in failures under unpredicted stress conditions; failures may also be maintenance-induced. Unlike the laboratory, field data collection systems sometimes fail to document the association(s) between failures and their cause or to "flag" maintenance difficulties. Therefore, they are not adequate to relate a particular problem to a specific design attribute, except in the most obvious cases. Nor do problems encountered in a small sample during a small portion of the expected life of a system necessarily accurately reflect what could be encountered with a larger sample or a longer sampling period. Both shortcomings preclude the development of design rules that would prevent such

problems. Yet translation of performance, costs, and schedule requirements into design rules and design features is critical to attaining the maximum advantage from the ULCE environment.

Application of computer techniques to the screening of data normally collected in the field in standard government format could reveal correlated patterns that, in turn, could be used to identify the cause of failure. These techniques must include safeguards to prevent incorrect conclusions (e.g., they should permit human interrogation and search subroutines).

It is essential that field failure data be added to data collected during laboratory tests (e.g., design qualification tests and prototype tests) to assess the difference between actual field experience and the tests with which the product was accepted for delivery. The comparison will serve two purposes: adjust both the field and laboratory data acquisition process and pertinent specifications for future programs; and develop design rules as well as laboratory test conditions that address the conditions experienced in the field.

It is important to associate the failure data with the specific serial number of the item being maintained, as well as to the next higher assembly and to the system itself (e.g., tail number). Such data would screen for trends that may be related to the next higher assembly or to the weapon system itself. The data collection system containing this information must also be capable of random access, so that comparison of the problems experienced with a specific item to those projected for that item can be made.

Dedicated products (e.g., maintenance aircraft) are clearly desirable and often essential for evaluation and verification of supportability. These supportability attributes under controlled conditions require dedicated products ranging from subassemblies to entire aircraft. This approach can be extremely costly because these dedicated products are usually not serviceable after a maintainability and supportability demonstration. However, development contracts do not always allow for such dedicated hardware. Therefore, the dedicated products are quite often mock-ups or a preproduction prototype that has undergone some other testing. Reduction of design data and credible supportability analyses may serve as a substitute for a large portion of this testing. It appears necessary to seek more cost-effective approaches to dedicated maintenance products—e.g., testing components rather than assemblies or coupling performance testing with simulation.

A costly portion of a maintainability and supportability demonstration is the evaluation of fault isolation capability. If the maintenance procedures require human intervention or interpretation of observations, as opposed to automated instructions, the demonstration is usually repeated with other technicians to obtain an average for performing the task. Designs should be sought that include appropriate features that would eliminate uncertainty and risk due to human interpretation. This type of design not only would be better from a supportability standpoint but also would reduce the cost of demonstrations, as well as reduce the risk of discovering problems during the demonstration, thereby preventing costly redesign.

A set of ULCE parameters for specific classes of applications needs to be established to track life-cycle impacts. The introduction of the term "supportability" in recent years has provided considerable confusion in the "ilities" communities. The term itself is made up of design attributes such as reliability, maintainability, and testability. Each of these measures of merit has

many well-defined quantifiers that are used by the different Services. It is essential that uniform quantifiers be agreed on or developed to simplify interpretation and application of data collected by the different Services to the preparation of design aids.

There will always be variations in cause-and-effect relationships. Some of these relationships may be between material characterizations and failure modes and frequencies, but others may be between design and failure information. It is not adequate to treat such variations uniformly with statistical methods because variations may be drastically different and have different implications for material selection and design. Advanced methods for processing uncertain and unreliable information will be needed (extended Bayesian statistics, fuzzy set theory, pattern recognition, machine learning, artificial neural nets). Information may appear to be uncertain or unreliable for a number of reasons, including the presence of nonessential related "noise," incomplete specification of process, or erroneous procedures for combining or generalizing interpretations.

In ULCE, a considerable amount of information will have much variation in it. This is especially so for the behavior of materials. There is a need to disentangle the extent to which such variations are caused by unreliable data collection or reporting procedures, variations in material characteristics and/or design and manufacturing factors. There is also a need for procedures for generalizing from such uncertain information to provide estimates of the projected behavior of new materials. A risk model is needed that permits assessment of the risk/benefit of applying a new, poorly characterized material rather than an old, well-characterized material. Also, a managerial tool integrating the life-cycle cost calculator (discussed under Critical Issue 2) with performance and operating criteria is required; this tool would be used to assess the risk/benefit that will be experienced in applying such material with shorter development and test cycles. That assessment should then be used to support investments in developing further analysis tools to study and project the behavior of new material. To develop a practical risk analysis, the relationship of cost drivers to the uncertainties of applying new materials must be understood and quantified in a manner that would permit rigorous mathematical analysis. A risk assessment capability is driven particularly by

• Application of new materials that incur long development cycles that include testing the materials' behavior during manufacturing processes and in the intended use environment, and

• Field data collection, as well as laboratory data, that only provide information on existing material and/or structures from which conclusions are drawn as to the material's failure mechanisms.

This latter approach has been found highly inaccurate in electronics for projecting failures of new components, since the underlying mechanism of their failure is not well understood. Failure mechanisms in structural materials, particularly the new ceramics and composite materials, are also poorly understood. Prediction of failure mechanisms is essential to developing designs that will survive application and maintenance as well as to designing built-in tests capable of projecting failures.

The use of sensor technology would solve a considerable amount of fault-detection and fault-isolation problems. This technology requires further development to make it applicable to weapon systems. For example, work at the Air Force Wright Research and Development Center's Material Laboratory in the area of inspection will potentially result in analytical tools that could evaluate a design to ascertain whether or not it can be inspected with techniques such as x-ray and eddy current measurements. This would prevent the need for demonstrating these design attributes with costly mock-ups.

There are a number of NDE techniques that offer the possibility of being utilized via embedded sensors to continuously monitor vital materials characteristics. These techniques include: fiber optic sensors utilizing fluorescent spectroscopy or polarization spectroscopy; dielectric or electrical conductivity measurement utilizing microwaves, capacitance probes, embedded electrodes or eddy current coils, and acoustic transducers with conventional electrical connectors or with an optical fiber array coupled to acoustic emission sensors. These techniques might be coupled to such conventional techniques as optical holography.

The relation of maintenance problems experienced to the design attributes that could have potentially caused these problems is normally studied under laboratory conditions. Laboratory conditions do not, however, provide an appropriate cross section of field technician understanding and capability and therefore fall short of the objective to facilitate maintenance as envisioned in the Air Force project Forecast II. Thus, there is increasing attention being devoted to techniques that combine hardware and software to automatically generate prototype parts from computer models of the part. At least three such techniques are under development and may have the potential to be useful in prototyping a broad spectrum of parts. In one, Nova Automation Corporation uses a laser to fuse layer upon layer of a powder into a preprogrammed shape. 3-D Systems, Inc,. uses a laser to photochemically cure liquid plastic in a process known as stereolithography. Again, numerous passes of the laser build layer upon layer until a complete part has been solidified. A third process, being developed by Hydronetics Inc., uses the laser pass technique to cut layers from metal stock, usually with a copper coating. The coating is used to bond each layer of the stock through a melting procedure.

A "human factors" feature of rapid prototyping techniques currently used for the development of software can be applied to ascertaining the proper execution of maintenance instructions. This feature would capture and relate the technicians' reactions to each and every maintenance instruction provided by their computers. Subsequent analyses could separate their lack of comprehension from design or instruction problems. At present, such separation would have to be performed manually, but artificial intelligence techniques could be designed to mimic at least the major portions of such screening. The rapid growth of the artificial intelligence field, as well as that of training aids, holds promise that automatic characterization of maintenance actions can be accomplished.

Techniques need to be developed to assist in ranking and selecting the critical parameters to assist the design engineer in addressing the essential issues in damage-tolerant design features. Artificial intelligence techniques can be used to assist in screening the many cause-and-effect

relationships of damage to material and aircraft structure from a design features standpoint, as well as the impact of such damage to reliability and supportability. This screening would assist the design engineer in addressing the most important parameters first. To support such artificial intelligence techniques, a data base will also need to be developed from actual field experience with aircraft damage, which in turn would relate such damage to reliability impact and aircraft availability.

Expert systems and new data bases currently being advanced should be developed in such a fashion that design knowledge and design rules can be easily sorted by dominant effect on items such as performance, reliability, manufacturability, and influence on cost and schedule.

The construction of this comprehensive data base would require a systematic approach that first investigates the needed knowledge in each step of a design process, how this knowledge is presented to the designer and design checker, and how it is utilized by them. If this data base were to be constructed to permit growth, it could be expanded to include the best design strategies and application of techniques and technologies collected from the best of evolving designs over the next decade. Human factors techniques will have to be applied extensively, so that the resulting design aids will be easy to use and readily accepted by the ULCE team.

Systems should be developed that can incorporate uncertain and unreliable information (e.g., material variability). To develop such a system, the progress being made in materials science should be coupled with classic stochastic finite element analyses as well as statistical techniques (such as variance analyses and reduction for Monte Carlo predictions) to serve as tools with which to perform the risk/benefit analysis of new materials application. Life-cycle cost modeling must be included in the risk evaluation to account for all characteristics that impact cost.

Classical Bayesian statistics continue to serve as a departure point for dealing with uncertainties. In that approach, probability is a purely objective measure based on frequency of occurrence, and methods for evaluating joint and conditional probabilities are well understood and can be verified experimentally.

The formalism of fuzzy set theory provides a significant departure from, and augmentation of, that practice by providing a framework for working with uncertainties that are not only inevitable but may, in fact, be desirable. This is because logic rules formulated in terms of fuzzy sets are much more inclusive and powerful than rules dealing with narrowly defined entities. Membership functions provide the interface between generalized statements and quantitative measurements. The former are understandable to humans and easy to store and retrieve, whereas the latter can be very specific and can be attained with use of well-specified procedures.

The disentangling of cause and effect may be aided with use of one or more of the autonomous machine-learning procedures under development. The ID3 method advocated by Quinlan (1983) and the rule inference procedures of Michalski (1980), Winston (1975), and Pao and Hu (1985), provide methods for learning from examples. Generalization procedures are important. Otherwise, the rules would be narrowly defined and not of general utility.

The rapidly developing artificial neural-net technology is also very useful in "discovering" underlying relationships between presumed causes and observed effects and can provide a

procedure for learning from a few examples to yield a powerful generalization mechanism. The mechanism is of such a nature that it cannot be described as a rule or even as a body of rules.

All these technologies are in states of active development, and efforts need to be made to determine how they might be utilized for processing of uncertain and possibly unreliable information in ULCE.

Colley, David P. 1988. "Instant Prototypes," Mechanical Engineering 110, No. 7, pp. 68–70.

Manufacturing Automation Protocol—MAP. 1987. NTIS, Springfield, VA. Citations from the Inspec: Information Services for the Physics and Engineering Database. Rept. for 1975.

Michalski, R. S. 1980. Pattern Recognition as Rule-Guided Inductive Inference, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol 2. No. 4 pp. 349–361.

Pao, Y. H., and C. H. Hu. 1985. Processing of pattern based information, Part 1: Inductive methods suitable for use in patten recognition and artificial intelligence. J. T. Tou, ed. Advances in Information Systems Science, Vo. 9, Plenum Press, New York.

Quinlan, J. R. 1983. Learning efficient classification procedures and their application to chess end-games. R. S. Michalski, J. G. Carbonell and T. M. Mitchell, eds. Machine Learning, Tioga, Palo Alto, CA.

Sowa, J. R. 1984. Conceptual Structures, Addison-Wesley, Reading, MA.

Winston, P. H. 1975. Learning structural description from examples. P. H. Winston, Ed. The Psychology of Computer Vision, McGraw-Hill, New York.