Approach
The project aims to address the need for flexible and efficient manufacturing that produces zero waste and high-quality products. To achieve this, COGNIMAN will start with a first phase that involves collecting and analysing the framework conditions of the selected manufacturing scenarios. This includes defining the context in terms of legal, procedural, ethical and technical challenges for a flexible smart manufacturing system, as well as specifying the key components required for an AI-enhanced robot system.
The project will then establish a human-centric technological design by elaborating on the different capabilities that will be developed for each pilot. These capabilities include Digital Twin (DT) and simulations, advanced sensing, machine learning and robots for cognitive manufacturing. It will integrate and pre-test these different components in a representative environment before conducting an extensive field trial at the manufacturing plant.
Cognitive Digital Twin and simulations
We plan to expand the COGNITWIN Toolbox for deploying DTs in manufacturing and robotics. The pipeline includes sensing, knowledge, reasoning, and action steps. The Toolbox manages data, provides semantic maps, and uses simulation libraries based on Robot Operating System (ROS).
Advanced sensing
We aim to establish a multimodal sensing system for cognitive manufacturing and robotics. Three sensor suites are defined for four diverse pilots.
Glass fibre pilot: Includes high-resolution camera, laser projection, near-infrared cameras, gamma measurement, ultrasonic sensors, radar and acoustic emission sensors.
Precision machining and additive manufacturing pilot: Includes time-of-flight cameras, confocal chromatic sensing, ellipsometry, interferometry and structural light imaging.
Steel manufacturing pilot: Requires sensors for personnel safety monitoring, product identification, storage space identification, safe flight and navigation of drones in indoor harsh environments.
Machine learning for manufacturing
We aim to create a machine-learning toolbox for tasks like fibre break detection, classification and safety prediction. It combines deep neural networks and graphical models for efficient sensor analysis. The project focuses on handling large datasets and enabling on-board AI for low-power computing. Challenges include real-time interpretation of portable hardware. Solutions include reduced network precision, smaller networks and new convolution approaches.
Robotics in manufacturing
Robot guidance for deburring/additive manufacturing: Optimal path planning for deburring/additive manufacturing using 3D maps and vision-based navigation. Continuous learning and safety measures ensure high-quality workmanship.
Drone for autonomous detection in metal production: Multi-drone solution for steel production warehousing, optimising logistics through RFID-based detection and active area mapping.
Collaborative robotics: Safe and efficient collaboration between robots, humans and machines. Long-term SLAM system and shared-awareness framework enhance performance.
Human centric design
COGNIMAN aims for human-centric smart manufacturing by exploring human-machine-robot teaming and developing guidance procedures. These key research questions will be investigated: optimal human-AI relationship and principles of interaction and achieving explainable, trustworthy and ethical AI. The project utilises human-in-the-loop AI learning for improved training and system output.