About

Background

Early-stage industrial design increasingly involves complex systems that combine software, hardware, robotics, cyber-physical components, and cybersecurity requirements. These design phases are often based on ill-defined problems, where textual requirements, functional descriptions, numerical data, and heterogeneous technical information must be interpreted together.

Despite the growing complexity of these systems, model generation, reasoning, and design flaw detection still rely heavily on human expertise and manual analysis. This creates risks of late-stage errors, costly redesign, software vulnerabilities, and unsafe system behavior.

The REINFORCE project addresses this challenge by developing a novel neuro-symbolic AI framework for automatic model generation and early detection of industrial design flaws. The project uses directed coloured graphs to represent system architecture, behaviour, and interactions, enabling computational reasoning over complex design problems.

REINFORCE builds on long-term research on qualitative reasoning and the DACM model developed at Tampere University. This foundation is combined with AI expertise from the University of Helsinki and cybersecurity expertise from the University of Jyväskylä. The approach integrates graph-based problem formulation, theory-driven neural networks, and inconsistency resolution methods inspired by TRIZ inventive principles.

Aim & Objectives

The project aims to advance early-stage industrial design towards automated, explainable, and fail-safe model generation.

The project has two main objectives:

1. To develop an automated model-generation framework based on directed coloured graphs and theory-driven neural networks, enabling simulation, system monitoring, hypothesis testing, system control, and design optimization.
2. To support the early detection and correction of design flaws, vulnerabilities, and inconsistencies in complex industrial systems, with applications in cyber-physical system design, robotic control, explainable machine learning, and cybersecurity-by-design.

Implementation

The REINFORCE project will develop a computational assistant for developers, designers, engineers, and decision-makers. This assistant will structure and analyse complex design problems, generate formal models from early-stage design information, and support decision-making before costly implementation choices are fixed.

The project will combine qualitative reasoning, dimensional analysis, graph-based representations, and neuro-symbolic AI methods. These methods will be used to automatically identify relationships between system variables, detect inconsistencies, and propose more robust design alternatives.

The framework will be applied to several industrially relevant domains, including explainable and parsimonious machine learning models, cyber-physical systems, robot system control, and early-stage cybersecurity assessment. By addressing critical issues during the early stages of the design process, REINFORCE aims to improve system quality, reliability, safety, and security in future industrial applications.