In this paper, we propose a domain-speci c modeling language for specifying the design of cyber-physical systems. The proposed domain-speci c modeling language can capture the control, computing, and communication aspects of a cyber-physical system design in an integrated manner. The concrete syntax of the proposed domainspeci c modeling language has been implemented as an extension of standard blocks available in Simulink. The meta-model for the proposed domain-speci c modeling language has been de ned using the Ecore meta-modeling language, which was originally developed as a part of Eclipse Modeling Framework project. We have also implemented an initial version of a parser that converts Simulink models, employing the proposed domain-speci c modeling language, into corresponding Eclipse Modeling Framework instance models, which can then serve as input to model transformation tools. The proposed domain-speci c modeling language builds on the concepts introduced by Giotto programming paradigm for platform-independent speci cation of control law, implemented by the controller in a cyber-physical system.
Embedded control systems consist of a physical plant being controlled by a feedback control algorithm, executing on an embedded computing platform. Typical examples of embedded control systems include automotive and avionics. Development of embedded control systems typically involves the following steps: requirements speci cation, control design and analysis, embedded design and analysis, and implementation. Model-driven development paradigm has been successfully applied to the eld of embedded control systems through the de nition of appropriate domain-speci c modeling languages (DSMLs) for di erent development steps of embedded control systems and appropriate model transformations among these DSMLs. Figure 1 shows the DSML and analysis tools used for each development step of embedded control system.
Last couple of decades have witnessed a dramatic decrease in the cost of sensing, communication, and computation technologies. This phenomenon has enabled the development of a new breed of embedded control systems that are much larger in scale, resulting in non-guaranteed wide-area communication subsystems. Moreover, these systems have longer lifespans and are more open in nature, resulting in the need for online recon guration and maintenance. Some examples of this new breed of embedded control systems are smart electric grids, smart irrigation networks, and vehicular networks. Design techniques and tools available in the domain of traditional embedded control systems cannot be directly applied to this new breed of embedded control systems. The eld of cyberphysical systems aims to address these issues by rethinking the abstractions involved in the development of embedded control systems. In particular, the eld of cyber-physical systems emphasizes the co-design of control, communication, and computation aspects of a system. As a result, the development steps for cyberphysical systems are requirements speci cation, cyber-physical system design (incorporating the co-design of control and computation), and implementation.
Figure 2 shows the development steps of a cyber-physical system (CPS) and analysis tools required at each of these steps. However, in order to develop a cyber-physical system according to model-driven development paradigm, the DSMLs used to specify the design of embedded control systems are not su cient. An appropriate DSML for CPS design speci cation must be able to capture the control, communication, and computation aspects of a CPS design. Moreover, an appropriate DSML for CPS design speci cation must independently convey the characteristics of a CPS computing platform as well as the control law implemented on that computing platform. The meta-model for this DSML can then be used to de ne model transformations that convert a CPS design specication into appropriate input for analysis tools (such as simulation and formal veri cation tools) as well as executable code for implementation platforms.
In this paper, we propose a DSML for CPS design speci cation. The
proposed DSML is capable of capturing the control, communication, and
computation aspects of a CPS design, as it provides blocks to represent physical system,
cyber system, and cyber-physical interface involved in a cyber-physical system.
The proposed DSML builds on the concepts introduced by Giotto programming
paradigm [
The rest of the paper is organized as follows. In Section 2, we present a brief review of the component technologies involved in the development of DSML, proposed in this paper. In Section 3, we present the details of the proposed
DSML for CPS design speci cation. In Section 4, we outline some related work. In Section 5, we present the conclusion. 2
In this section, we brie y review the technologies involved in the de nition and implementation of the proposed DSML for CPS design speci cation. 2.1
Simulink, developed by MathWorks, is a simulation and model-based design tool that provides a graphical editor for specifying a model as a set of hierarchical block diagrams. Simulink is often used in conjunction with some auxiliary tools that provide specialized types of blocks to be used in Simulink block diagram. Two important examples of such auxiliary tools are State ow and Simscape. State ow allows the users to model decision logic based on the state machine and ow chart formalisms. Simscape provides fundamental building blocks from various domains (such as electrical, mechanical, and hydraulic) that can be combined to model a physical plant.
Simulink has become a defacto standard in the eld of embedded control systems. It is not only used to simulate and analyze the proposed designs, but various model transformations have been de ned to generate executable code from Simulink models for various embedded platforms. Therefore, we have chosen to develop the concrete syntax of the proposed DSML for CPS design speci cation as an extension of the standard blocks available in Simulink toolset. 2.2
Eclipse is an open-source software project, aimed at providing a platform that
can be reused for the development of integrated development environments
(IDEs). Eclipse is divided into numerous top-level projects such as Eclipse Project,
Modeling Project, Tools Project, and Technology Project [
At the core of the Eclipse Modeling Project is Eclipse Modeling Framework (EMF) and all the other model-driven development technologies (such as modelto-model transformations and model-to-text transformations) have been built on top of EMF. At it core, EMF is a framework for de ning a model and generating Java code from that model. As a part of EMF, Ecore modeling language has been de ned that can be used to specify the model from which EMF generates a set of Java classes and interfaces. However, in the model-driven development context, an Ecore model can also serve as the meta-model of a DSML. We have used an Ecore-based meta-model to specify the abstract syntax of our proposed DSML for CPS design speci cation. 2.3
Typical development process for an embedded control system requires a
collaboration between control engineer and software engineer. First, a control engineer
models the physical plant, derives the feedback control law, and validates the
controller design through mathematical analysis and simulation. Then, a
software engineer decomposes the computational activities of a feedback controller
into time-constrained software tasks, develops code for these tasks in a
traditional programming language (such as C), and assigns priorities to these tasks
so that they could meet their timing constraints while being scheduled by an
RTOS scheduler. Giotto is a specialized programming language for embedded
control systems that aims to bridge the communication gap between a control
engineer and a software engineer by providing an intermediate level of
abstraction between control design and software implementation [
Giotto program provides a platform-independent description of feedback
controller design in terms of time triggered sensor readings, actuator updates, task
invocations, and mode transitions. Then, a Giotto compiler compiles the Giotto
program onto a speci c computing platform, preserving the functionality as well
as the timing behavior, speci ed by Giotto program. Any CPS design speci
cation must independently convey the computing platform characteristics as well
as the controller design that needs to be implemented on that computing
platform so that this design speci cation can then serve as an input to appropriate
CPS analysis tools. Therefore, our proposed DSML builds on the concepts
introduced by Giotto programming paradigm for platform-independent speci cation
of the control law, implemented by the controller in a cyber-physical system.
Figure 4 reviews the major concepts involved in a Giotto program [
In this section, we present the details of the proposed DSML for CPS design speci cation. The individual elements of the proposed DSML can be divided into three categories: physical system blocks, cyber system blocks, and cyber-physical interface blocks. CompoundPhysicalPlant, AtomicPhysicalPlant, PhysicalSystemParameter and PhysicalLink blocks belong to the physical system blocks. Sensor and Actuator blocks make up the cyber-physical interface blocks. ComputingNode, NetworkLink, NetworkRouter, NodePlatform, NodeApplication, Mode, Task, ModeSwitchLogic, SensorPort, ActuatorPort, InputMsgPort, and OutputMsgPort make up the cyber system blocks.
Physical system component of a CPS design can be speci ed by a set of AtomicPhysicalPlant blocks connected to each other through PhysicalLink blocks. PhysicalLink blocks represent \ ow of energy" between components of a physical plant. A set of AtomicPhysicalPlant and PhysicalLink blocks can also be grouped together into a CompuondPhysicalPlant block. Moreover, PhysicalSystemParameter blocks are used to identify the elements of a physical plant that are to be sensed and actuated upon by the cyber system. Cyber-physical interface of a CPS design is captured by a set of Sensor and Actuator blocks. Each Sensor and Actuator block is connected to a PhysicalSystemParameter block.
Cyber aspects of a CPS design include the network topology of computing nodes, platform characteristics of each computing node, and the application software executing on each computing node. These aspects can be captured by connecting a set of ComputingNode, NetworkRouter, and NetworkLink blocks. Each ComputingNode block must include a NodePlatform and NodeApplication block. NodePlatform block in turn includes Processor, RTOS, Middleware, and ApplicationFramework blocks, which allow it the capture the platform characteristics of a computing node. The NodeApplication block includes a set of Giotto-inspired blocks that can specify the feedback control algorithm executing on that computing node in a platform independent manner.
Concrete syntax of the proposed DSML has been implemented as an extension to standard blocks available in Simulink. In particular, a new Simulink library has been developed that provides a Simulink block for each of the proposed DSML element. We have also used Simulink's mask interface capability to provide each new Simulink block with a custom look, and a dialog box for entering element-speci c parameters, such as the bandwidth and delay of NetworkLink block. Figure 5 shows a Simulink model that speci es a CPS design using the proposed DSML. Figure 6 shows the internal details of CompoundPhysicalPlant block. Figure 7 shows the internal details of a ComputingNode block, which contains a NodeApplication and a NodePlatform block. Figure 8 shows the internal details of NodeApplication block, which consists of a set of Mode blocks and a ModeSwitchLogic block. Figure 9 shows the internal details of Mode block, which contains a set of Task blocks. Figure 10 shows the simpli ed version of the Ecore-based meta-model for the proposed DSML. 4
Integration of Simulink with Giotto has also been pursued in [
In [
Generic Modeling Environment (GME) is a popular domain-speci c
modeling environment [
In this paper, we have proposed a DSML for CPS design speci cation that is capable of capturing the control, computing, and communication aspects of a CPS design. The proposed DSML leverages the concepts of Giotto programming paradigm to ensure that the CPS design speci cation independently conveys the CPS computing platform characteristics as well as the control law that needs to be implemented on that computing platform. The abstract syntax of the proposed DSML has been de ned through an Ecore-based meta-model. The concrete syntax of the proposed DSML has been implemented as an extension of standard blocks available in Simulink. We have also implemented an initial version of a parser that converts Simulink models, employing the proposed DSML, into corresponding EMF instance models, which can then serve as input to EMF-based model transformation tools.
The individual elements of the DSML, proposed in this paper, will be re ned further in the future based on the experience obtained from di erent case studies. However, the proposed DSML with the choice of EMF for meta-modeling and Simulink for concrete syntax can serve as the basis of integrating various emerging facets of CPS research such as CPS simulation tools, CPS formal veri cation tools and CPS implementation platforms.