Hybrid Approach to Intelligent Recommenders for Testbed DevOps



Starting point and vision: The automotive industry has long been at the forefront of
advances in automation. Modern assembly lines and supply chains are almost completely automated. Recent advances in data science and digitization have led to intelligent systems where data across the heterogeneous system landscape is becoming a major source of innovation, as was equally the role of innovative functions in the past. The complex data-driven discipline of vehicle development is supported by complex cyber-physical systems of systems (CPSoS) manifested by vehicle testbeds. However, novel propulsion systems are forcing the automotive industry to act faster than ever before. Development efficiency is the key and includes not only rapid development, but also the economic use of resources. HybridAIR aims to significantly increase the efficiency of CPSoS operation by introducing intelligent systems in the form of self-adaptive explainable recommender systems for different stakeholders applied on testbeds.

Problem to solve: How can we cope with a constantly increasing volume of data and
data exchange between a multi-stakeholder and heterogeneous system landscape
(H2H, H2M, M2M)? How can we profitably turn around the associated challenges to
create a source of knowledge from the many (informal) data sources, which not only
mitigates the increasing complexity but even further increases efficiency? More
concrete, how can we provide explainable, stakeholder-specific, high-quality, and
context-aware decision support for testbed developers and operators (DevOps),
based on existing, shared data sources enriched with contextual information in their
respective fields? How to build a generic, reusable infrastructure portable to reach a
broad target group, whose respective knowledge is profitably reinforced?

Motivation of R&D project: Tackling this issue requires HybridAIR to develop a noveland original mix of methods, techniques and tools based on, to this date, unrelated research fields. These include research on multi-stakeholder recommender systems (MSRS), where human-centred design will be the key, alongside the aggregation and fusion of data with online and offline contextual information. Online information by stakeholder observations together with AI-based approaches to enable adaptive, context-aware systems should not contradict trustability by providing
recommendation explanations adapted to the respective stakeholder (XAI/AXRS).

Degree of innovation: HybridAIR will contribute with approaches to enhance existingdata representations to enable their semantic (XAI, DSL) and contextual processing. With the aimed multi-stakeholder approach, knowledge networks, discovery ofcross-stakeholder relations and automation of knowledge transfer between stakeholders will become possible.

Planned results: The main result will thus be corresponding integrated cognitivesystems in the form of multi-stakeholder and context-adaptive recommender
systems (MSRS) based on explainable AI (AXRS). They will be capable of modelling
human mental reasoning for different involved roles such as testbed operations and
development (DevOps). Advanced human-computer interfaces (e.g., AR) and
stakeholder-specific tooling will support decision-making and field observations to
ensure sustainable growth of a shared multi-stakeholder knowledge base.
StatusNicht begonnen


  • Wirtschaftsuniversität (Leitung)
  • AVL List GmbH (Projektpartner*in)
  • Technische Universität Graz (Projektpartner*in)
  • GUEP Software GmbH (Projektpartner*in)