DESNA

The DESNA research commercialisation project was funded by Enterprise Ireland between 2008 and 2009 and investigated the development of tool support necessary to facilitate the construction of sensor-network applications.

Motivation

Relatively little work has been done on appropriate and complete development environments and tool support to facilitate the easy development of sensor-network applications for different sensor-network platforms and different business domains that range from healthcare to environmental monitoring.

To alleviate the complexity of designing for diverse resource-constrained sensor platforms while being able to easily express domain concepts effectively, development tools have to employ high-level domain-specific languages to describe the collaborative behaviour of sensor nodes. In order to support rapid application development and the intelligent processing of sensor data, the development environment has to allow the integration of software components (such as Artificial Intelligence algorithms) from pre-exisitng component libraries, as well as provide a highly-customisable and scalable environment for realistic simulation of the resulting sensor-network applications prior to their deployment.

Scientific Basis

In order to alleviate the complexity of designing complex and heterogeneous sensor-network applications DESNA provides developers with high-level domain-specific languages and a toolset for:

Profiling the sensor-network platform (Domain Modelling),
Deploying customised planning, optimisation and inference activities from a library of software components implementing machine learning algorithms (Artificial Intelligence Modelling),
Expressing the control- and data-flow of sensor-network applications as business processes using the customised activities (Business Process Modelling), and
Performing realistic simulations of the resulting sensor-network applications prior to deployment (Simulation).

Results

DESNA provides a four-step process for the development of sensor-network applications as intelligent context-aware workflows. DESNA provides a methodology, algorithms and tools to assist developers in the process of application development, which can then be simulated using the DESNA simulation tool.

Fig. 1. A screenshot of the DESNA
domain modelling tool.

Domain Modelling

A key step in the DESNA methodology is the domain modelling step. This step is used by domain experts to create and populate domain models that capture relevant information on the infrastructure in a region. This multi-layered data model chiefly consists of data on the physical infrastructure and sensor and actuator networks in the region that provide support to the system being developed. An example of relevant physical infrastructure would be road and transport network data in traffic applications.

Artificial Intelligence Modelling

During the project we undertook the development of three distinct classes of AI algorithms in the context of two use case scenarios. We have developed and integrated a Bayesian Network that is used for inference purposes, a Simulated Annealing algorithm used for multi-attribute optimisation, and a regression model based on the Cox proportional-hazards regression methodology. These algorithms were developed in the context of the CCTV Monitoring and Vineyard Harvesting case studies. The figure below shows the Bayesian network defined in the CCTV monitoring case study.

Business Process Modelling

DESNA supports the definition of context-aware workflows as Business Process Execution Language (BPEL) processes in a service-oriented computing environment. BPEL allows business process developers to quickly define the high-level business workflow of sensor-network applications. Through BPEL one may easily integrate data sources (e.g., sensor-network data streams, historical data), AI widgets and tools (e.g., the DESNA simulator) as WSDL services that the business process orchestrates. The DESNA Business modelling tool provides a visual way to quickly design, validate, deploy and execute context-aware intelligent business processes, that is, BPEL processes that integrate AI widgets and sensor-data streams. The figure below presents a screenshot of the DESNA business modelling tool.

Simulation

Using the DESNA simulation tool the user can quickly model the deployment environment for the application whose functionality needs to be tested. The user defines different phenomena in terms of state layers and assigns initial values for each phenomenon with respect to the simulated time scale. These initial values infer the evolution of each phenomenon separately. The user can also define events for each phenomenon which will be triggered at specific points in simulated time as configured by the user. These events are defined in terms of Javascript code, which gives the user complete freedom to manipulate phenomenon for which the event is defined and other phenomena present in the simulation in the form of state layers. The simulator also allows the user to define inter-state layer dependencies in terms of Javascript code. These dependencies represent the dependencies between different phenomena being modelled in the simulation and are executed at each simulation iteration. The figure below shows a screenshot of the DESNA simulation tool.