Project Outline and Objectives
AI-PROFICIENT (Artificial Intelligence for improved PROduction efFICIEncy, quality and maiNTenance) is a thirty-six month Horizon 2020 project funded by the European Commission, launched in 2020, aspiring to integrate advanced Artificial Intelligence (AI) technologies with manufacturing ones, aiming to improve production planning and execution and facilitate the collaboration between humans and machines.
In order to face challenges of adaptability, agility and resilience of Cyber Physical Production Systems, the project team investigated disruptive technological solutions, and introduced AI for the implementation of optimised strategies in the frame of agile production and predictive maintenance. Also, AI-PROFICIENT project places the human factor at the centre of the implementation of these solutions, by considering ethical aspects in relation to the work of each operational actor. This way, it upraises Ethics at the core of the service design by adopting an Ethics-by-design approach.
The project addressed a series of Digital challenges within the modern factories. In an effort to achieve smart component integration and interoperability, an Industrial Internet of Things (IIoT) installation was the basis to collect data from the shop floor and provide it to a series of higher-level modules implementing the solution.
Smart components were deployed for embedded AI at system edge, AI detection and diagnostics and prognostics for system degradation and health state assessment.
The overall tool collection comprises:
- Semantic lifting,
- Model agnostic techniques used for Explainable AI (XAI),
- Process Modelling done by using Hybrid Digital Twins,
- Generative optimisation of production processes,
- AI-enabled decision making,
all provided in an AI Platform, facilitating Process Optimisation and also, Quality Assurance.
Comprehensive Human-Machine Interfaces (HMIs) were created to visualise output originating from all developed tools, using role-specific visualisation for machine operators, providing tools for transparent AI decision support. Thus, usage of the HMI applications, deployed in project pilot sites, accommodates the operator with a Plant Management System putting Human-In-The-Loop.
Project Use Cases, Tools developed and Applications deployed
AI-PROFICIENT was demonstrated in three operation environments (i.e. production facilities), tackling different challenges, under different scenarios of significant economic value, which were then made available by two manufacturing enterprises acting as consortium members.
- Continental, a tyre manufacturing with Combiline system for rubber preparation, extrusion and cutting in the Sarreguemines plant (France);
CONTINENTAL Sarreguemines site
- Ineos, producing petrochemicals with two plant facilities for production of polypropylene and polyethylene polymers in Geel (Belgium) and Cologne (Germany) respectively.
Pretty Pictures
INEOS Geel site
INEOS Cologne site
Potential ethical and legal obstacles were carefully studied as part of the demo production facilities and adequate recommendations were provided and disseminated in this regard.
The tools developed by the project are the following:
- Surrogate data driven model based on explainable AI
- Connected worker Additive Check application
- Quality Data analysis: GO-QData
- Machine vision for dimensional measurement and positioning with small “bad” datasets
- Data-driven predictive AI analytics for prognostics based on DL
- Natural Human-Computer Interaction
- Process anomaly detection: GO-QNormality
- Multi-objective optimiser (generative and MI(N)LP optimisation solvers)
- Semantic Accountability Tool
- Short-term Post-hoc Anomaly Detection tool
These tools are implemented into two remarkable HMI applications. The applications deployed, each exemplifying the transformative power of innovation, are the following:
INEOS Application: An Augmented Reality (AR) enabled mobile application that allows for new user interaction paradigms. By seamlessly integrating voice-controlled commands with Quick Response (QR) code scanning, the application empowers users to engage with their surroundings in a hands-free manner. This synthesis of AR, voice recognition, and real-time data processing exemplifies the transformative potential of immersive technologies in enhancing operational efficiency.
Ineos Application
Continental Dashboard: A web application displaying information from a sophisticated system designed for real-time AI-driven machinery monitoring, and operator feedback. The web application not only presents operators with essential machinery data but also offers AI-generated suggestions for value adjustments. This interactive loop, wherein operators’ responses inform AI model refinement, showcases the collaborative synergy between human expertise and algorithmic insights.
Continental Dashboard
Outcomes achieved
Based on tools implementation within the Industrial environment of Continental and Ineos, important AI-based improvements were realised both at shop floor level operation and at production quality. Namely, the project achieved optimisation of machine control setting and set-up of product parameters, leading to significant reduction of production failures due to system breakage/disturbance, which consequently leads to minimizing low quality products due to system degradation. Improved model operation due to data drift identification combined with operator-on-the-loop had a further contribution on increasing product quality. At the same time, the use of intelligent image recognition optimised the usage of raw materials whilst AI-based operations achieved increased raw material conversion.
Project unique points and innovations moving EU Industry forward
The project has developed a human-centered (anthropocentric) approach to AI, rather than a technology-centered one. Ethics by design is therefore at the heart of the project, as a goal and not as a constraint. In addition, AI-based technological solutions were developed taking into account the particularities of the application field (manufacturing), the data scarcity (e.g. new equipment, small data sets), the data uncertainty (e.g. problem of sensor settings), the representativeness (e.g. normal operation, degraded operation), the dynamics of the manufacturing systems (e.g. old age, wear and tear) etc. Furthermore, the provided AI solutions can be combined on different abstraction levels (Edge, Fog, Cloud, Decision) of a manufacturing system to provide anticipation (e.g. field-level intelligence), proactivity (e.g. information management) and overall performance. Finally, the AI-based solutions were tested in an operating environment with real conditions (e.g. CONTINENTAL), enabling the measurement of industrial credibility/benefits and scalability under severe conditions.
Who is behind this project
AI-PROFICIENT was made possible through the collaboration of ten partners from six European countries all of them having a distinct role in the project. Here is a short summary of the partners, including the technological and scientific offerings.
- Universite de Lorraine (France): Coordinator of the project, worked on System Health, self-diagnostics/self-prognostics, predictive AI-Analytics and Proactive maintenance strategies. Also, worked on Ethics-by-design and supported standardisation activities.
- Continental France SNC (France): Provided industrial scale pilot installation and Use Cases to develop problem solving solutions.
- Tekniker (Spain): Worked on field-level automation, self-diagnostics, production process anomaly detection, Predictive AI analytics.
- Ineos Services (Belgium): Provided two industrial scale pilot installations in two different EU countries, and provided three Use Cases to develop and demonstrate problem solving solutions.
- Tenforce (Belgium): Worked on Semantics, Role-specific human-machine interfaces, XAI and Conversational interfaces at shop floor level.
- VTT (Finland): Worked on Digital Twins and Hybrid models describing production processes.
- Inos Hellas (Greece): Worked on component level data acquisition and pre-processing, on proactive maintenance strategies in the Industrial environment and leaded Standardisation activities.
- Ibermatica (Spain): Worked on IIoT environment deployment/setup, and Use Case Validation methodology and Analysis.
- Institut Mihajlo Pupin (Serbia): Worked on IIoT interoperability, Generative optimisation, product-level anomaly detection, XAI and transparent AI decision making, System architecture and Platform deployment.
- Athens Technology Center (Greece): Worked on Optical character recognition (OCR) and Speech-to-text technologies in production environment and on machine/process-level Anomaly detection, followed by Root-cause analysis and AI reinforcement learning algorithms. Leaded also the dissemination, communication, exploitation and clustering activities of the project.
For more information on the AI-PROFICIENT project, visit: https://ai-proficient.eu/