Smart IoT systems, or SIS for short, create new automation possibilities, e. g. for intelligent buildings, digital health applications and efficient transportation systems. Here, they interact with other devices or humans.  Due to this, the development must meet high demands in terms of security, reliability and data protection. In addition, SIS have to cope with very dynamic environments, whose conditionsh are difficult to predict at the time of development. The SIS must be able to adapt independently to changes in the environment, e. g.  to data security threats occurring during runtime. This requires new approaches to the creation and operation of these systems. 

In the EU project ENACT, Prof. Pohl’s paluno working group, together with European research partners, has further developed the process improvement approach “DevOps” especially for SIS. The term DevOps is made up of the words development and IT operations and aims to interlink these areas, which are actually working separately. For this purpose, the project partners designed suitable software tools for all phases of the DevOps lifecycle (see graphic). 

Artificial intelligence can support DevOps

The working group of Prof. Pohl was responsible for the design and prototypical implementation of the “Online Learning Enabler”, which supports the “Operate” phase with automated tools. “We investigated how artificial intelligence enables the run-time adaptation of IoT systems and tested different, innovative learning algorithms”, explains project coordinator Dr. Andreas Metzger. “We were able to achieve the best results in terms of automation with policy-based reinforcement learning. With these algorithms, the IoT software is able to learn appropriate adaptations for different and changing environmental situations.” 

But what happens if the functionalities of the SIS are changed during manual development or maintenance of the software? “The learning algorithms are initially blind to such changes”, explains Dr. Metzger. “In order to enable them to take account of newly introduced adaptation possibilities through further development, we use variability models. These models, known from software product line development, enable adaptations to be presented and analyzed in a compact and machine-readable form.” 

The work of the ENACT partners adds up to an integrated DevOps framework for common IoT platforms. The project was funded by the European Union for a period of three years with a total of 4.93 million euros. 420,000 euros of this went to the UDE. The results are summarized in the recently published and freely accessible book "DevOps for Trustworthy Smart IoT Systems".