Industrial Domains
Manufacturing
The Industrial Domain Manufacturing is applying AI to optimize operations, enhance automation, and improve resource utilization. Key uses include predictive maintenance, AI-powered robotics, production planning, and real-time supply chain monitoring, all while ensuring compliance with safety, environmental, and ethical regulations. Challenges include AI integration with existing systems, data privacy, and addressing ethical concerns like job displacement and algorithmic bias. Adaptive AI systems help boost agility and resilience in manufacturing processes.
Project: Self X-Smart MAintenance for industrial Robot manipulator Technologies (X-SMART)
Third parties involved: Simone Arena – University of Cagliari
The X-SMART project (self X-Smart MAintenance for industrial Robot manipulator Technologies) aims to improve knowledge of predictive maintenance (PdM) strategies for industrial robot manipulators by integrating Self-X capabilities (Self-Configuration, Self-Healing, Self-Optimization, Self- Protection). Key activities include developing a Cyber-Physical System tomonitor performance and applying adaptive AI for real-time fault detection. Expected results include downtime reduction, improved system adaptability to unforeseen events, and the development of adaptable, autonomous PdM software. The research will contribute to companies’ assets availability, competitiveness, and sustainability, by advancing trusted and adaptive AI practices.
Project: Improving low data volume defect detection via neural architecture search
Third parties involved: Fain Tech doo
This project applies neural architecture search (NAS) to tackle challenges posed by low-volume datasets, while pairing the results with explainable AI (XAI) to ensure interpretability. Building on our recent breakthroughs, we will begin by establishing solid baselines, then progressively advance toward the core NAS and XAI objectives. Drawing on prior experience, we anticipate strong performance gains from NAS. The work aligns with European AI priorities, pushing the state of the art in AI-driven manufacturing and promoting reliable, trustworthy AI systems through robust explainability methods.
Project: LEarning Defect Detection from Samples with REduced size and SYnthetiC Labelled data (LeddsResycl)
Third parties involved: Joanneum Research
LeddsResycl proposes to address the detection of components of interest and their classification into defect-free or defected in two separate steps. Building on the following strands of deep learning work: One is few-shot detection/classification using pairwise fine-grained training, able to perform incremental training from limited sample sets. The other is the model soup paradigm for training instance segmentation networks on synthetic data, and improving generalisation to real data. We further will include our expertise in explainable AI for classification in order to provide semantically meaningful feedback to users of the AI system. The proposed methods will be evaluated against approaches from literature.
Project: AI-Driven Predictive Maintenance and Risk Management Approach for Machine Tools
Third parties involved: Hakob Grigoryan – NVISION
This study proposes a novel approach to predictive maintenance and health monitoring in machine tools by integrating Failure Mode Effect Analysis (FMEA) with advanced Deep Learning algorithms. Predicting and assessing machine breakdowns is crucial for optimizing maintenance schedules, reducing downtime, enhancing operational efficiency, and minimizing financial losses associated with failures. While traditional rule-based and systematic methods like FMEA offer valuable insights into potential failure modes, they often lack real-time predictive capabilities. In contrast, AI algorithms can analyze large datasets to identify patterns indicative of impending failures. By combining FMEA with AI, this research aims to develop a comprehensive framework for proactive maintenance that accurately identifies and predicts potential machine breakdowns.
Project: Self X-Smart MAintenance for industrial Robot manipulator Technologies (X-SMART)
Third parties involved: Simone Arena, University of Cagliari
The X-SMART project (self X-Smart Maintenance for Industrial Robot Manipulator Technologies) aims to improve knowledge of predictive maintenance (PdM) strategies for industrial robot manipulators by integrating Self-X capabilities (Self-Configuration, Self-Healing, Self-Optimisation, Self- Protection). Key activities include developing a Cyber-Physical System to monitor performance and applying adaptive AI for real-time fault detection. Expected results include downtime reduction, improved system adaptability to unforeseen events, and the development of adaptable, autonomous PdM software. The research will contribute to companies’ asset availability, competitiveness, and sustainability by advancing trusted and adaptive AI practices.
Project: Improving low data volume defect detection via neural architecture search
Third parties involved: Igor Jovančević, Nikola Milović, Nikola Pižurica – Fain Tech doo
This project applies neural architecture search (NAS) to tackle challenges posed by low-volume datasets, while pairing the results with explainable AI (XAI) to ensure interpretability. Building on our recent breakthroughs, we will begin by establishing solid baselines, then progressively advance toward the core NAS and XAI objectives. Drawing on prior experience, we anticipate strong performance gains from NAS. The work aligns with European AI priorities, pushing the state of the art in AI-driven manufacturing and promoting reliable, trustworthy AI systems through robust explainability methods.
Project: LEarning Defect Detection from Samples with REduced size and SYnthetiC Labelled data (LeddsResycl)
Third parties involved: Joanneum Research
LeddsResycl proposes to address the detection of components of interest and their classification into defect-free or defected in two separate steps. Building on the following strands of deep learning work: One is few-shot detection/classification using pairwise fine-grained training, able to perform incremental training from limited sample sets. The other is the model soup paradigm for training instance segmentation networks on synthetic data, and improving generalisation to real data. We further will include our expertise in explainable AI for classification in order to provide semantically meaningful feedback to users of the AI system. The proposed methods will be evaluated against approaches from literature.
Project: StressLens: Integrated XR & AI System for Industrial Stress and Safety
Third parties involved: LTG and ESOGU-IFARLAB-EDIH
StressLens project develops an AI-based system using EDA and eye-tracking data via XR headsets to monitor worker stress and deliver real-time safety alerts. It focuses on personalised stress detection, prevention of cognitive overload, and real-time feedback. It includes multimodal sensing, adaptive model training (e.g., CNN, SVM), and validation in XR-based industrial testbeds. Aiming for over 90% accuracy in stress prediction, it enhances safety and worker well-being. It provides strong EU added value by enabling safer, more efficient workplaces through scalable, ethical, and open-source AI. Aligned with ENFIELD goals, it advances human-centric, trustworthy, and adaptive AI via GDPR-compliant, individualised monitoring and dynamic alerts.
Project: FOCUS: FOot-based Control for User Support
Third parties involved: Logimade
FOCUS aims to develop and validate an AI-based foot gesture recognition system to enhance human–machine interaction in waste sorting stations. Operating in harsh environments, workers wear extensive PPE, limiting traditional interfaces. FOCUS enables hands-free control of conveyor speed, waste feed rate, and container replacement, while supporting real-time communication along the sorting line. With the support of Madeira’s largest municipalities and ARM, and building on Logimade’s existing service infrastructure provided to these entities, the system will be tested in real-world facilities. The project will improve operational efficiency, safety, and inclusiveness, contributing to Europe’s digital and green transitions.
Project: SoleSense: AI-Based Real-Time Foot Gesture Control for HMI
Third parties involved: Sensifai BV
Industrial operators need hands-free control without removing protective gear. SoleSense is the first on-device, low-latency foot-gesture Human-Machine Interface (HMI) with sub-100 ms response and >85% accuracy across eight co-designed gestures (e.g., “heel tap,” “toe press”). Co-design is a collaborative methodology where end-users actively shape the lexicon. A green-AI CNN, trained on 50,000+ synthetic foot poses, outperforms open-source estimators under occlusion and eliminates insole hardware. Scalable to RGB cameras, SoleSense reduces workflow interruptions by 25%, improves accessibility for gloved users, and accelerates adoption in AMRs, cobots, and heavy machinery.
Project: Self X-Smart MAintenance for industrial Robot manipulator Technologies (X-SMART)
Third parties involved: Simone Arena, University of Cagliari
The X-SMART project (self X-Smart Maintenance for Industrial Robot Manipulator Technologies) aims to improve knowledge of predictive maintenance (PdM) strategies for industrial robot manipulators by integrating Self-X capabilities (Self-Configuration, Self-Healing, Self-Optimisation, Self- Protection). Key activities include developing a Cyber-Physical System to monitor performance and applying adaptive AI for real-time fault detection. Expected results include downtime reduction, improved system adaptability to unforeseen events, and the development of adaptable, autonomous PdM software. The research will contribute to companies’ asset availability, competitiveness, and sustainability by advancing trusted and adaptive AI practices.
Project: ASTRID: Advanced Software Tools for Reliable Industrial Datasets
Third parties involved: SCCH – Software Competence Center Hagenberg GmbH
The Advanced Software Tools for Reliable Industrial Datasets (ASTRID) project develops open-source tools and metrics to evaluate and improve dataset reliability for AI in manufacturing. The project addresses bias, information leakage, and robustness by defining quality checks for sensor logs, images, and synthetic data. Activities include collecting real and synthetic datasets, designing new reliability metrics, and validating software in real-world settings. Results will be openly released, supporting safer and more trustworthy AI adoption in the European industry.