Design and Evolution of Cyber Physical Systems: A Dynamic Data Driven Application System

Abstract

Design of Cyber Physical Systems is a complex and challenging endeavor. Cyber Physical Systems by definition span a broad spectrum of design domains -- cyber and multi-physics; design abstractions -- hybrid dynamics, ODE's, PDE's, solid geometry, among others; and multiple design disciplines and tools. The design has to satisfy a large number of, often conflicting, requirements spanning performance, structural, manufacturability, reliability, maintainability, and cost. Current systems engineering methodologies rely largely on the traditional approach commonly known as the System V method. The V method follows a top-down decomposition of design into subsystems, partitioning requirements along well established disciplines and teams, while going down the V. The subsystems are brought together for integration and testing at a later phase while going up the V. The challenge arises from unanticipated interactions that are discovered during integration, requiring costly design iterations. The result of this has been an untenable trend-line of cost and schedule vs complexity for military and aerospace systems. Augustine postulated, based on this trend-line that, without a revolutionary design approach, cost and schedule will put new globally competitive military systems out of budgetary reach.We present the OpenMETA toolchain that was created in DARPA's AVM program to address the challenges of complex Cyber Physical Systems design articulated above. OpenMETA implements a design methodology that can be characterized as progressive constraint-based design space refinement, and relies on three core principles, as described below: component-based design, design space exploration, and automated requirement driven analysis. Component-Based Design supports design reuse, leveraging well-designed component models, configured in architectures designed to achieve system requirements. Design space models allow representation of flexible design decisions, with associated Design Space Exploration tools to find sets of feasible designs. Design spaces allow significant design adaptability, a major goal of AVM, and also minimize design churns that are caused by requirement creep traditionally necessitating retaining and carrying forward a large design space through progressive refinement. Instead of changing a single point design as requirements evolve, in the OpenMETA methodology the design space is refined, typically through sub-setting. Executable Requirements support automated evaluation of system metrics across entire feasible design spaces, significantly reducing the cost of system validation. When used with the supported probabilistic methods, functional correctness can be assured within the defined probabilistic bounds that account for the multiple sources of uncertainty in the design process, including but not limited to model uncertainty. This OpenMETA design flow is implemented with a comprehensive set of tools, editors, and frameworks that maximally leverage available best-of-class commercial and open-source engineering tools, while reducing the expertise in each of these tools via abstraction and design automation. The toolset is built upon Meta Programmable Tools and a Semantic Backplane that support extensibility and semantically precise integration of domains and tools.In OpenMETA, design decisions are made based on models of the components and systems. A significant challenge remains however, with respect to the validity and fidelity of the models. The Dynamic Data Driven application approach represents an opportunity to address this challenge. The DDDAS methods and infrastructure, could allow us to close loop between the system design process (models and tools) and the operational system. Instrumentation and observations from the operational system can be used as a feedback to tune the design models to match the realworld observed behavior. The refined models, in turn, can be used to optimize the system. In this workshop, we will present an early exploration of techniques for model adaptation and evolution that are being investigated for integration in the OpenMETA toolchain.