Modeling Methodology for Smart Manufacturing

Summary

The key to efficient and effective manufacturing operations is oftentimes knowledge of what will and will not work. In most instances, this knowledge may be gained through analytical and empirical methods. Currently however, the tasks of developing analytical and empirical models and translating results into recommendations are costly and error prone. Fundamental challenges to using analytical and empirical methods include making use of information from various sources, knowing that the techniques to be applied are appropriate to the situation, knowing the extent to which analytical results are valid, and acting on the insight the effort provided. More so than in engineering design, the preparation of models and the interpretation of results in manufacturing do not follow a precise methodology shared across various usages. This limits the ability to develop systematic means to apply analytical and empirical techniques to manufacturing decision making. The consequences of this include missed opportunities to reuse knowledge, unreliable results, and high cost of analysis. This project will develop and demonstrate new ways of applying analytical and empirical methods using domain-specific modeling methodologies tailored for smart manufacturing. This research will enable analytical frameworks that support our methodology. These frameworks will bring analytical capability to smart manufacturing, reducing the cost of manufacturing and increasing the speed by which smart manufacturing systems adapt to changing requirements and external events.

Description

Objective:  Develop new modeling methodologies, guidelines, and software tools to enable manufacturers to more easily apply analytical methods to manufacturing operations and processes.

Technical Idea:  Manufacturers would realize great benefit were the costs of composing analytical models reduced. Currently, the models and information sources used in analytical activities are not easily integrated toward accomplishing specific manufacturing objectives. A principal barrier to integration is a lack of methods of model composition. Though substantial research has been performed in this area in the context of systems engineering, little of it takes advantage of the unique characteristics of smart manufacturing. These characteristics include on-going need for analytical methods (in contrast to analytical methods focused on the design process), integration of data from operations, integration with production control systems, and the ability to dedicate a portion of manufacturing resources to experimental investigation of new processes.

The variety of problems to which analytical methods may be applied makes specifying a single integration strategy difficult. Nonetheless, there exists a common thread among the strategies we address. Domain-specific modeling languages is an emerging technology that provides an efficient means to represent a variety of viewpoints, but lacks a methodology for effective composition in manufacturing settings. The project will develop methodologies for the representation and composition of analytical models synthesizing elements of domain-specific modeling languages. The project will develop prototype analytical frameworks to exercise these methodologies.

Research Plan:  We have identified four new model composition strategies relevant to smart manufacturing systems that will extend the body of research we believe is needed to bring new analytical methods to solve complex manufacturing problems. We are developing frameworks (including domain-specific methodologies and tools) to facilitate integration in the four strategies. The four strategies are:

1. A component-based strategy: In this strategy, an analysis is conceived as the integration of components from libraries of model elements. The goals of our work here are to (a) produce a methodology and a tool for the composition and optimization of process chains, (b) demonstrate ontology-based type checking the flows between model elements, and (c) demonstrate a process to produce surrogate models from these base models. The work will produce reports describing the research and a publicly-available pilot tool.
2. A Semantic Web-based strategy: In this strategy, an analysis is developed through queries against information organized as relational triples (subject-predicate-object). Conversely, triples may be derived from content in disparate viewpoints (e.g., a process description viewpoint, an equipment characteristics viewpoint). The work will produce reports describing the research and a publicly-available pilot tool.
3. A tiered-architecture strategy: In this strategy, an analysis is formulated from information organized by metamodels (models of modeling languages) of two types: domain-specific metamodels and solution technology metamodels. Domain-specific metamodels describe viewpoints on manufacturing information (e.g., a process plan viewpoint, a product demand viewpoint, a production resource viewpoint). Tool technology metamodels describe an interface to the capabilities of an analytical tool. Model composition in this context requires bridging the gap between the domain-specific and tool-technology tiers. The work will be demonstrated in concert with participation in the development of ISO-15746, a standard for integration of advanced process control and optimization capabilities for manufacturing systems. The work will produce a metamodel of optimization which may be promoted for standardization and reports describing the research.
4. A system identification strategy: In this strategy, a grey-box model of system behavior in the form of stochastic Petri nets is discovered through evolutionary algorithms. The intended purpose of this model is to interpret data and system behavior. The work will produce a pilot tool and reports describing the research.