As organizations seek to improve their competitive position by responding effectively to the increasing rate of change in the market place, the need for agile enterprises and agile manufacturing has come to the forefront. In this paper we examine the essential role that information sharing plays in enabling the agile manufacturing of complex products. The principle goal of an agile manufacturing interface is to provide the information sharing infrastructure necessary to enable the formation of virtual organizations and to provide them with the robust DFx (Design for x, where x = producibility, testability, maintainability, etc.) mechanisms they need in order to develop high quality products in a timely, cost effective manner. Results achieved from the implementation of an agile manufacturing interface in a production environment are highlighted. These results include a 10x reduction in the cycle-time required to go from design to manufacturing set-up, and a reduction in the rework for complex Printed Circuit Assemblies of up to 80%.
1. Introduction
In order to remain globally competitive, it is critical that organizations establish processes that allow them to adapt to an ever changing market place. This need for agility is particularly pronounced for organizations involved in the design and/or manufacture of complex products. A manufacturing interface that allows design and manufacturing organizations to interact in an effective and efficient manner is essential to realizing the necessary agility. Such an agile manufacturing interface must allow designers to quickly asses the level of compatibility between a design and a manufacturing facility, while simultaneously providing manufacturers with the ability to effectively interact with a diverse variety of design organizations.
The Rapid Prototyping of Application Specific Signal Processors (RASSP) Program is a $150M ARPA and US Department of Defense initiative (1994-1997) intended to dramatically improve the way complex embedded digital electronic systems, particularly embedded digital signal processors, are designed, manufactured, upgraded, and supported. The target RASSP improvement is at least a four-fold (4x) reduction in the time to go from design concept to fielded prototype with equivalent improvements in cost and quality. The motivation for the RASSP initiative is the pervasive need for affordable embedded signal processors throughout a wide range of electronic systems.
To achieve the RASSP 4x objective, the RASSP program has supported the development of an agile manufacturing interface. The RASSP Manufacturing Interface provides the mechanisms necessary for diverse design and manufacturing organizations to work synergistically and thereby help ensure first-pass manufacturing success.
In this article, the information sharing and DFx requirements for an agile manufacturing interface are presented. The principle goal of an agile manufacturing interface is to provide the information sharing infrastructure necessary to enable the formation of virtual organizations and to provide them with the robust DFx mechanisms they need in order to develop high quality products in a timely, cost effective manner. The results achieved from the implementation of an agile manufacturing interface to support the RASSP process are presented. These results, realized in a production environment, include a 10x reduction in the time required to transition from design to manufacturing set-up, and a reduction in rework of complex Printed Circuit Assemblies (PCA) of up to 80%. The role that effective information exchange plays in enabling these results is highlighted.
2. Background
A traditional flexible manufacturing system will provide some of the capabilities needed to achieve both the reductions in cycle-time and cost and the improvements in quality sought by the RASSP program. However, stronger coupling between design and manufacturing is needed to fully realize the RASSP objective. To achieve this, an agile manufacturing interface must enable close collaboration between design and manufacturing groups throughout the product development process. An agile manufacturing interface supports this level of collaboration by providing the information sharing infrastructure and DFx mechanisms necessary to develop complex products that achieve first-pass manufacturing success.
2.1 Limitations of Traditional Approaches
Today, the information sharing capabilities and DFx mechanisms necessary to support the sophisticated collaboration requirements of agile enterprises do not exist. Typically, little communication occurs between organizations until critical information exchange is required. When an exchange of information occurs, little or no consideration is given to the receiving group’s information requirements. This results in what is popularly referred to as the over-the-wall paradigm of information exchange.
PCA design and manufacturing organizations all too often operate within this paradigm. The PCA design group will declare a design “finished” once it meets their requirements. Typically, these requirements focus on form, fit, and function while ignoring other considerations such as producibility. It has long been recognized that such a paradigm is inefficient. Companies have taken steps to adopt concurrent engineering principles to ensure that other considerations, such as manufacturability, are taken into account earlier in product design. Several different approaches have been used in an attempt to accomplish this. In the electronics area, one approach has been to incorporate generic Design For Manufacturing (DFM) rules into Electrical CAD (ECAD) systems. ECAD systems equipped with this primitive DFM capability can analyze a PCA and report issues that might negatively impact producibility. A second approach has been to form Integrated Product Development (IPD) Teams consisting of representatives from various disciplines other than design, such as manufacturing. These domain experts are typically collocated with the design team to ensure that their domain’s concerns are considered during the design process. Therefore, by including a manufacturing engineer on the IPD team, manufacturing knowledge specific to one facility can be applied to improve the producibility characteristics of product designs.
Both of these approaches have improved the overall situation. However, each has significant drawbacks. Neither approach enables the consideration of multiple manufacturing facilities or production lines, nor do they support the need for agility that is so important in today’s defense and commercial environments. In order to be as effective as possible, DFM rules must include knowledge specific to the manufacturing facility and production line that will be used to produce the product being designed. Every manufacturing facility will contain different equipment, will organize its equipment differently, and will have its own idiosyncrasies. These idiosyncrasies can significantly impact producibility and are often poorly documented, if documented at all. Knowledge of this kind is generally distributed among the technicians and manufacturing engineers that operate a manufacturing facility. Because these idiosyncrasies are not considered by the generic DFM checking capabilities that ECAD systems provide, serious producibility issues can exist after the generic DFM rules indicate that no issues remain. Because unknown producibility issues often exist when manufacturing begins, first-pass manufacturing success is rarely achieved.
While the IPD team approach can significantly improve the producibility of designs, it is far from an ideal solution. First, resource limitations constrain the number of manufacturing personnel that can participate on an IPD team, ensuring that only a subset of a manufacturing facility’s characteristics can be taken into consideration by the design team. Second, physical collocation of manufacturing experts with designers is expensive and subject to potential interpersonal management issues. Third, the manufacturing knowledge used with this approach exists primarily in the mind of the manufacturing engineer, and thus is volatile from the design organization’s point of view. Access to this knowledge can be interrupted or eliminated by factors such as illness, changes in employment, and retirement. Finally, the IPD approach does not easily support the optimization of a design across multiple manufacturing facilities. Because the field of potential manufacturers is severely restricted early in the design process, this approach restricts a design organizations flexibility.
What is needed is an agile manufacturing interface that provides the mechanisms necessary to enable an automated, concurrent engineering environment. Such a solution must eliminate the fundamental, underlying impediments to first-pass manufacturing success of complex products, allow design organizations to quickly interface with different manufacturing facilities, and simultaneously allow manufacturing facilities to effectively interface with many different design organizations [Gad94].
2.2 The ARPA/Tri-Service RASSP Program
The ARPA/Tri-Service Rapid Prototyping of Application Specific Signal Processors (RASSP) program is a 4.5 year, $150M effort aimed at improving the process by which embedded digital electronic systems are developed. The objective of the RASSP program is to reduce by a factor of four the cost and time needed to develop and manufacture embedded signal processing systems while simultaneously improving their quality. RASSP has targeted three areas for development in support of achieving this objective:
Methodology
Model Year Architecture
Infrastructure
The methodology being developed combines concurrent engineering concepts with collaborative teaming approaches. The model year architecture focuses on leveraging commercially available capabilities, coupled with flexible interfaces, to enable regular, low-cost technology upgrades. Improvements in infrastructure are being pursued to increase the effectiveness of the methodology and model year architecture being developed. These infrastructure efforts are focused on two areas. The first is aimed at improving the capability of system level design tools that can be used to automate and improve the decisions made early in the design process. The second focus area is developing improved enterprise integration capabilities, such as enterprise product data management (PDM) systems integrated with workflow management systems and enterprise reuse libraries. By providing integrated workflow management and secure, high bandwidth Internet access, the infrastructure effort will enable application of the methodology and model year architecture across distributed, multi-discipline concurrent engineering teams within a virtual corporation.
A critical component of the enterprise integration capabilities being developed by the RASSP program is the RASSP Manufacturing Interface (RASSP-MI). The goal of the RASSP-MI is to enable first-pass manufacturing success of PCAs within a virtual enterprise by effectively supporting agile manufacturing. This goal directly supports RASSP’s goal of significantly improving the quality of and reducing the time and cost required to design and deploy signal processor systems.
3. Agile Manufacturing Interface Requirements
In order to enable the formation of virtual organizations, an agile manufacturing interface must provide a robust information sharing infrastructure coupled with the DFx mechanisms necessary to realize cost effective, first-pass manufacturing success. The following sections discuss these requirements in greater detail and describe how they have been implemented in the RASSP Manufacturing Interface.
The information sharing requirements for an agile manufacturing interface can be divided into two categories. The first requirement is that the information being shared have a predictable and mutually agreeable form; that is, its syntax must be understood by both sender and receiver prior to any exchange of information. The second requirement is that the information being shared have a predictable and mutually agreeable content; that is, the semantics of the information to be exchanged must be understood by both sender and receiver prior to any exchange of information. This implies that data “flavoring” is not allowed. The content or semantic requirement can be further refined as follows:
Semantic Requirements
Complete
Consistent
Accurate
Correct
When the data requirements of a receiving activity, such as manufacturing, are not formally understood by a sending activity, such as design, the completeness of the information transferred can not be assured. For example, features of a design considered insignificant to a designer may be crucial to a manufacturer. A design organization may transfer what they consider to be a complete design to the manufacturer, only to discover later that the manufacturer requires more information. Rectifying this situation requires a costly, time consuming iteration between design and manufacturing before production can begin.
The consistency of the information transferred from one activity to another must also be assured. The absence of Integrated Product Data Management (IPDM) between organizations results in an environment in which the consistency of the information may be compromised. For example, it is not uncommon for a manufacturing engineer to make a “minor” change to the layout of a design, such as changing a signal name in the netlist, after the layout has been completed. The resulting lack of consistency between different views of a product can result in costly and unnecessary delays in manufacturing. These delays will result either from problems caused directly by the inconsistent information (when the inconsistency goes unrecognized), or due to the design iterations required to correct the inconsistency.
The most challenging requirements to meet are those of accuracy and semantic correctness. The issue of accuracy arises when information is exchanged in a form other than its native representation. Indeed, even exchanging information in native form can present accuracy problems unless the environment for both sender and receiver (architecture, software environment, etc.) are identical. A design is correct when it meets its functional specification and all “ility” requirements are satisfied, such as manufacturability, testability, maintainability, etc. The need to verify that these “ility” requirements are met drives the need for robust DFx checking mechanisms that perform design validation as part of an agile manufacturing interface.
4. The RASSP Manufacturing Interface
Figure 1 presents the RASSP Manufacturing Interface (RASSP-MI) architecture. The role played by each component of the RASSP-MI architecture in realizing the agile manufacturing interface requirements described above is presented in the following sections.
4.1 Information Sharing Standards
The utility of standards in a concurrent engineering environment can be seen in [And94] which describes the role standards played in supporting the concurrent engineering of the engine mount for the Boeing 777 aircraft.
The RASSP-MI makes effective use of robust, widely accepted standards to provide “data buses” which ensure that information can be exchanged between product life-cycle elements (design, manufacturing, testing, field service, etc.) without the loss or duplication of information. The role for information sharing standards in the context of the RASSP agile manufacturing interface is illustrated in Figure 2.
The use of robust standards in the RASSP-MI supports many of the information sharing requirements described previously. Standards such as EDIF [Lau96] and ISO 10303 (STEP) [ISO96] are specified such that they meet the predictable form requirement and can support the content requirements in several ways. By enabling the representation of all needed information, the completeness requirement is supported. The consistency requirement is supported by defining rules as part of the standard that can be used to automatically check the consistency of the information to be exchanged. Lastly, by providing a formal definition of the semantics of the information to be exchanged via an information model, both EDIF and ISO 10303 enable techniques that can ensure the information exchanged is accurate. This is discussed in more detail next.
4.2 Assuring Product Data Accuracy
To ensure that the information exchanged accurately conveys the semantics of the original representation, the RASSP-MI makes use of a novel EXPRESS driven approach to data conversion. Using this technique, a formalized definition of the information’s semantics is created using the EXPRESS information modeling language [ISO94]. This formal definition enables semantic mappings between different representations to be developed. An example of a semantic mapping is illustrated in Figure 3.
Note that this mapping is incomplete; a complete mapping of the PIN entity would show the correspondence between all attributes in the two information models. Using the EXPRESS Driven Data Conversion technique defined in [Hin94], these semantic mappings form the basis for accurately converting data from one form into another.
4.3 Assuring Product Data Correctness
As described earlier, the most challenging information sharing requirement, that of semantic correctness, can only be met by robust DFx mechanisms. This drives the need for robust DFx checking mechanisms within the RASSP agile manufacturing interface.
The DFx capability supports the validation of a data-set’s correctness relative to specific criteria, such as producibility and testability. The DFx mechanism applies rules to check a data-set against the specified criteria. The values and semantics of the specified criteria exist as a machine processable description of relevant manufacturing capabilities in terms that are meaningful to a designer. An example of a DFx rule is presented below:
Example DFx rule:
" traces . if trace_width < min_trace_width
=> issue ( min_trace_width )
In the above example, trace_width and min_trace_width represent variables. The variable trace_width is determined from the description of a product, whereas the variable min_trace_width is obtained from the process description that represents a manufacturing facility’s capabilities. The DFx analysis mechanism will evaluate the above rule to determine if any trace in the design is narrower than the minimum trace width defined by the manufacturing facility. If a trace is found that violates this condition, an issue is generated that can be resolved through negotiations between design and manufacturing to either alter the manufacturing process, alter the design, or ignore the issue. The manufacturing-facility-specific DFx analysis supports an iterative cycle of design analysis and refinement, which can be repeated until no significant issues remain for a design. While some non-fatal issues may be unresolvable due to design constraints, knowledge of these will result in more realistic manufacturing cost estimates than could be achieved without the aid of such DFx capabilities.
The RASSP-MI provides the necessary DFx capabilities through a World Wide Web (WWW) accessible Producibility Analysis (PA) tool. Together, the WWW and PA support the secure transmission of design information, remote analysis, the secure return of analysis results, and any ensuing negotiations that may be required. An example of the information provided by the RASSP-MI PA tool is presented in Figure 4.
The architecture described here has been used to develop the agile RASSP-MI. The results obtained while using the RASSP-MI in a production environment are presented next.
5. Results To Date
The RASSP-MI has been integrated into the RASSP enterprise system and is being utilized by the key PCA manufacturing facility within Lockheed Martin Corporation. Several PCA designs have been processed by the RASSP-MI at this facility to produce a number of PCAs. The results to date indicate a significant reduction in rework and design-to-manufacturing cycle-time. These results are detailed next.
5.1 Baseline Production Environment
Until recently, the process of transitioning PCA product designs from Lockheed Martin’s design facilities to its key manufacturing facility involved significant manual data conversion, data reentry, and manual quality assurance procedures. These manual processes required significant time to perform and introduced errors and inaccuracies into data generated for production. These data conversion and quality assurance steps took place after a PCA design was considered “complete” and had been transferred to the manufacturing facility.
Because the manufacturing facility has traditionally not been part of the product design process, manufacturability issues are often present in data received from design. These issues must be resolved before production can begin. Resolution might require a re-design effort by the team originating the design. Because the cost of design modification late in the design cycle is high, manufacturability issues that are not insurmountable are often allowed to remain, even though they increase the recurring manufacturing costs of the product. These problems have not only contributed to difficulty in achieving first-pass manufacturing success, but unnecessarily increased production difficulties and therefore cycle-time and cost.
The RASSP-MI corrects this by facilitating collaboration and negotiation between design and manufacturing engineers throughout the product design process. The role of the RASSP-MI in this process is illustrated in Figure 5.
Two practices are still in place at the manufacturing facility which are legacies of previous product data generation methods. The first of these practices is to manufacture the first three PCAs of the first manufacturing run of a new design by a purely manual process. This is done to ensure that the process of product assembly is well understood. This knowledge can then be used to help address difficulties encountered during automatic assembly of the remaining PCAs. The second legacy practice is to produce only 20 PCAs per manufacturing run (referred to as a batch). This is done to provide significant opportunities for manufacturing engineers to fine-tune the automatic assembly process in order to maximize yield.
Prior to use of the RASSP Manufacturing Interface, inaccurate placement of surface-mount components caused significant recurring production difficulty. The manual data exchange process employed did not assure accuracy of placement information to within 1/1000th of an inch. Without this level of accuracy, it was common for small discrete surface-mount components to move during the solder reflow process due to component drift. For some components, this movement caused them to make poor or no contact with their designated connection points on the Printed Circuit Board (PCB). Attempts to counter this effect centered around modifying “offset” values in the automatic surface-mount placement equipment. Failures observed during the manufacture of a batch of PCAs would be analyzed by a manufacturing engineer, who would then use the analysis results to modify placement equipment “offset” values in an attempt to correct the component misplacement problem. This approach improved yields, but was never able to eliminate this production problem, even over several years of production of the same design.
Despite the ingenuity and tenacity of the engineers and technicians supporting this facility, the inaccurate data utilized for production exacted a heavy toll. For one program examined, 100% of 80,000 manufactured PCAs had defects caused by inaccurate placement of surface-mount components. These defects required manual repair. To make matters worse, on average approximately 30% of the components on each PCA required rework. Remarkably, it was determined that the cost required to overcome these difficulties, given the over-the-wall paradigm the facility was obligated to operate within, exceeded the cost of performing the repairs.
5.2 Results Using the RASSP-MI
To date, four PCA designs have been processed using the RASSP-MI. These PCA designs are comparable in complexity to the design previously discussed. Using the RASSP-MI, NC code for component placement machines is derived automatically from the original CAD data representation of the design. Therefore, the placement information in the NC code is as accurate as that present in the CAD system. Due to the increased quality of the placement data, it was determined that all of the “offset” values that had been programmed into the surface-mount placement equipment at the manufacturing facility could be reset to 0, which resulted in a simplification of the programming procedures required for this equipment.
Of the four designs processed thus far, three realized first-pass manufacturing success. The remaining design experienced a 70% success rate. For this design, examination showed that a misinterpretation of the manufacturing facility’s information requirements was the cause of the poor yield. The EXPRESS driven approach to data conversion allowed this problem to be quickly identified and corrected.
In addition to supporting first-pass success, the RASSP-MI reduced the design to manufacturing set-up time by more than a factor of 10. The first-pass success and cycle-time improvements were achieved by adhering to the information sharing requirements described previously, eliminating unnecessary process steps, and providing an automated concurrent engineering capability between design and manufacturing.
5.3 Payback Analysis
Equation 1 below defines Ct to be the recurring cost associated with the time required to correct surface-mount component placement errors introduced by the baseline manual data conversion process previously described in section 5.1.
Where:
MElr is the labor rate of a Manufacturing Engineer
Tm is the time spent modifying automatic placement “offset” values per day of production
Pr is the percent of PCAs requiring repair due to poor component placement
Tr is the average time spent repairing a PCA
Np is the total number of PCAs produced
MTlr is the labor rate of a Manufacturing Technician
Using the RASSP-MI, Ct is negligible. Using the baseline process, Ct was significantly higher. Equation 2 presents the production savings on a per unit basis that has been enabled using the RASSP-MI, SRASSP.
Using Equation 1 and assuming typical fully burdened labor costs, the per unit savings enabled by the RASSP-MI are $20/PCA, as shown in Equation 2. Given the production rate of the manufacturing facility, the development costs of the RASSP-MI will be paid back in under 6 months.
It should be noted that significant quantities of the four PCA designs processed to date will be produced in the near future. Perhaps even more importantly, given the benefits identified through the use of the RASSP-MI, an additional 25 design projects are expected to be processed by the RASSP-MI over the coming months.
These results highlight the benefits of the agile RASSP Manufacturing Interface and explain why the Lockheed Martin PCA manufacturing facility was identified for a best-practice award [Best95]. With further refinements, it is expected that first-pass manufacturing success of PCAs will be consistently achieved using the capabilities provided by the RASSP agile manufacturing interface.
6. Summary
The goal of an agile manufacturing interface is to enable the formation of virtual organizations by providing the information sharing infrastructure and robust DFx mechanisms those organizations need in order to develop successful products. This paper presented the requirements for an agile manufacturing interface and the results obtained using the agile manufacturing interface developed by the RASSP program (the RASSP-MI) in a production environment. By reducing cost and time-to-market, the RASSP-MI is contributing significantly towards the accomplishment of the RASSP program’s goals of a 4x improvement in cycle-time, quality and cost.
In conclusion, the RASSP Manufacturing Interface allows physically distributed design and manufacturing teams to work collaboratively in a virtual organization to design manufacturability into complex products early in the design process. It also ensures that complex product designs are ready to be manufactured before production begins, thereby ensuring first-pass manufacturing success. For complex products in general, implementations of this capability promise to produce significant reductions in product development time and cost while improving product quality.
Acknowledgments
The research presented in this paper has been supported in part by the Defense Advanced Research Projects Agency (DARPA) Electronics Technology Office (ETO) and the Army Research Laboratory under the RASSP program, subcontract TTM 748358. Special thanks are extended to Ronald A. Pierce and David Dunham (Manufacturing Engineers) who provided invaluable assistance in collecting historical manufacturing facility information and results data.
References
[Gad94] A. J. Gadient, G.R. Graves & J.C. Boudreaux, “PreAmp: A STEP Based Concurrent Engineering Environment for Printed Circuit Assemblies,” In Proceedings Concurrent Engineering: Research and Applications Conference, pp. 529-537, August, 1994.
[And94] B. Anderson, S. Ryan, “Using STEP Application Protocols to Enable Concurrent Engineering in Real World Pilot Implementations”, In Proceedings Concurrent Engineering: Research and Applications Conference, pp. 349-353, August, 1994.
[Lau96] Lau, R.Y.W., EDIF: Electronic Design Interchange Format Version 4 0 0 Information Model, Electronic Industries Association, EDIF Steering Committee, 1996.
[ISO96] ISO/DIS 10303-210:1996, Industrial automation systems and integration - Product data representation and exchange - Part 210 Printed circuit assembly product design data.
[ISO94] ISO 10303-11:1994, Industrial automation systems and integration - Product data representation and exchange - Part 11: Description methods: The EXPRESS language reference manual.
[Hin94] L.E. Hines, A. J. Gadient, “EXPRESS Driven Data Conversion,” In Proceedings Concurrent Engineering: Research and Applications Conference, p. 313-322, August, 1994.
[Best95] “Report of Survey Conducted at Lockheed Martin Electronics & Missiles, Orlando, FL”, Best Manufacturing Practices Center of Excellence, College Park, Maryland, April 1995.
Anthony J. Gadient
Advanced Technology Group, SCRA
5300 International Blvd.
N. Charleston, SC 29418 gadient@scra.org
Lynwood E. Hines
Advanced Technology Group, SCRA
5300 International Blvd.
N. Charleston, SC 29418 hines@scra.org
