Conference Agenda

Product variants are designed to be configured under technical constraints, yet many configurators provide limited recommendations. This paper investigates support for product configuration by analyzing Recommender Systems and Large Language Models and their complementary use within the configuration process. Based on a literature review, a functional allocation along the configuration pipeline is derived to support hybrid recommendations based on product knowledge, historical configuration data and to enable natural-language interaction. These components are complemented into a conceptual framework that uses a formal constraint solver to consolidate technical feasibility.

Successful utilization of AI to support product development processes often depends on the early articulation of domain-specific AI use cases by domain experts, yet empirical insights into this procedure remain limited. This contribution presents an artifact-based analysis of 33 structured AI Idea Sheets applied by domain experts across different temporal and collaborative settings. The findings reveal strong problem and data specificity but systematic under-specification of AI framing and evaluation, indicating where interdisciplinary scaffolding may be required in early-stage AI system conception.

Design for X (DfX) methods are widely used to integrate manufacturability, performance, cost, and sustainability considerations into early design phases. However, in highly complex engineering systems, these requirements frequently contradict each other and are commonly evaluated in isolation using static decision schemes. This paper addresses this limitation by proposing a conceptual framework for resolving conflicting DfX requirements through dynamic, agent-based negotiation. A five-layer multi-agent system architecture is introduced, combining contextual requirement definition, graph-based design representation, domain-specific DfX agents, structured negotiation mechanisms, and knowledge integration with learning capabilities. Instead of statically scoring design alternatives, conflicting requirements are treated as negotiable entities that are iteratively reconciled by interacting agents. The approach is illustrated using the example of an additively manufactured high-pressure mini-channel heat exchanger, which exhibits tightly coupled thermomechanical and manufacturing constraints. While additive manufacturing serves as an exemplary application domain, the proposed concept is designed to be generalizable to other engineering fields characterized by high conflict density. The contribution establishes a foundation for adaptive, transparent, and learning-capable DfX decision support systems.

Integrating LLM-based agents into Model-Based Systems Engineering (MBSE) promises to reduce coordination overhead across heterogeneous toolchains and data silos, but unsystematic integration can result in agent sprawl, loss of human control, and opaque decision processes. This paper presents an event-driven reference architecture for Human–AI Collaboration in MBSE comprising three pillars: an Event-Driven Information Backbone (EIB) that combines pub/sub coordination with an immutable event log, Domain-Specific Agents (DSAs) that act as intelligent intermediaries between engineering tools and the backbone, and Personalized Interaction Agents (PIAs) that filter event streams into role-relevant decision packages and operationalize human-in-the-loop governance through a Decision Inbox pattern. All participants interact exclusively through explicit event contracts, enabling modular integration without bilateral coupling. Correlation identifiers link causally related events across tools and actors, enabling end-to-end reconstruction of decision provenance, including rejected alternatives and their rationale. A proof-of-concept prototype provides initial evidence of technical feasibility.

Generative AI (GenAI) is rapidly entering design practice still its concrete impact on human ideation (for example using it as inspirational stimuli and design fixation) remains insufficiently understood. This research investigates how GenAI influences creative performance and experience in early-stage design and how it can be integrated with traditional ideation techniques rather than using it as a stand-alone idea generator. The study employs both quantitative and Qualitative methods to collect data from a structured design exercise conducted with 14 students from a course on the use of AI in design creativity in university settings. Techniques such as Brainstorming, Brainwriting, Negative Ideation, and Prompt Engineering were employed in the design process to investigate how AI tools function as team members during collaborative design ideation. Creativity outcomes were evaluated on fluency, novelty, feasibility, surprise and then it was complemented by fixation scores. The results show that GenAI consistently increased perceived inspiration and helped participants reframe problems by providing new and fresh perspectives. The GenAI was widely experienced as an active “team-mate” rather than a simple tool. However, human-only conditions often produced comparable or higher idea fluency and AI-generated examples sometimes introduced new anchors that shifted the focus of design fixation rather than removing it. Negative Ideation coupled with decorated prompts, helped participants transform AI suggestions instead of copying them directly. The insights offer strategies for integrating GenAI into professional work and design education so that it grows without restricting the range of potential solutions.