CLEAR | Trustworthy multimodal AI for industrial applications

Region:

Europe

Financed by

Integrating heterogeneous industrial data to deliver more accurate and dependable AI

Industrial companies generate vast amounts of data from sensors, technical documents, images, 3D models and many other sources. Turning this heterogeneous information into reliable AI-driven insights remains a major challenge. CLEAR (Comprehensive Learning for Enhanced AI Responsiveness) develops trustworthy multimodal AI that enables Large Language Models (LLMs) and Large Multimodal Models (LMMs) to process complex industrial data more accurately, safely and efficiently.

Target group

The project results are relevant for organisations developing or deploying AI-driven systems that must process complex, heterogeneous data, including:

Industrial manufacturers and engineering service providers
Agricultural technology and precision farming organisations
Railway, transport and mobility operators
Telecommunications and network management providers
AI software developers and platform integrators
Research organisations working on responsible AI and Industry 4.0

Context

Large Language Models and Large Multimodal Models are transforming the way organisations interact with data. However, today's AI systems still struggle to process industrial information that goes beyond text, such as 3D models, sensor data, time series, technical documentation and geospatial information. They also remain vulnerable to hallucinations and inconsistent reasoning, limiting their adoption in safety-critical and highly regulated environments.

As AI becomes increasingly integrated into industrial processes, organisations need solutions that are more reliable, transparent and compliant with emerging European regulations such as the EU AI Act.

Objectives & results

CLEAR aims to develop trustworthy multimodal AI technologies that enable industrial organisations to integrate and exploit heterogeneous data more effectively.

The project will:

Develop new methods for combining unconventional multimodal data within LLM- and LMM-based systems
Build a modular AI pipeline covering knowledge integration, prompt engineering, data processing and response generation
Develop techniques for synthetic data generation, model fine-tuning, explainability and confidential AI
Improve the processing and alignment of time-series data for time-critical industrial applications
Validate the developed technologies across use cases in additive manufacturing, agriculture, railway and telecommunications
Support the adoption of trustworthy AI in line with the European AI regulatory framework

Approach

Within CLEAR, supported by VLAIO and the ITEA4 framework, Sirris works with Materialise on an additive manufacturing use case. Sirris brings its expertise in Generative AI, while Materialise contributes industrial know-how, real-world datasets and the application context.

The work focuses on two complementary research tracks.

1. AI-assisted quotation generation
Sirris develops AI methods that automate parts of the quotation process for additive manufacturing. The objective is to analyse customer requests, evaluate the printability of 3D models and generate draft quotations for human review.

2. Industrial knowledge retrieval
The project also investigates how AI agents can efficiently access and exploit large volumes of heterogeneous organisational knowledge. Sirris develops and evaluates advanced retrieval techniques, including semantic search and GraphRAG, to improve the accuracy and contextual relevance of AI-generated responses.

Together, these developments contribute to more reliable, explainable and trustworthy AI solutions for industrial applications.