Problems: CPS could be a breath sensor returning measurements to a mobile application, storing the data in the Cloud. Its data are massive, have different forms (e.g. physical measurements), and are treated in different ways by several software and devices (i.e. apps on mobile, Cloud systems, physical sensors).
Typically, data processed by most software systems is constrained by data models which are nothing more than types/concepts and well-formedness rules (e.g. a month is an integer, and is between 1 and 12). These simple specifications of the data domain are created due to the uncertain nature of unstructured data collected and processed by software systems, which is the case in CPS we are considering. However, the imprecise definition of the domain often postpones addressing important problems such as suboptimal memory usage (e.g. large domains require memory excessively to store data), lower sustainability (e.g. regression testing is computationally expensive in very large domains), and above all a risk of error due processing data of low quality outside the perceived operational limits of a CPS.
Objectives: Dizolo aims to involve the user (or worker) in the loop to maximally automate the design of constraints in CPS. We call this process design space exploration and the project will aim to achieve the following objectives over a period of one year.
O1: Development of a human-in-the-loop tool for incremental design of constraints from CPS data. The tool will combine human expert feedback with automation using formal methods such as Alloy,constraint programming, and association rule mining for both structured and unstructured data.
O2: Implementation and evaluation of the tool in two industrial cases: (a) A CPS for breathing developed by Norwegian Sweetzpot AS (b) A Cloud system used by the french PhD student.
O3: Leverage results from the collaboration to establish a larger European consortium addressing CPS for the H2020 work program factories of the future.