Periodic Reporting for period 2 - AI-PROFICIENT (Artificial Intelligence for improved PROduction efFICIEncy, quality and maiNTenance)
Okres sprawozdawczy: 2022-05-01 do 2023-10-31
Different sectors of manufacturing and process industry are characterized by the need for adaptive configuration of the control parameters and rapid response to fluctuating production demands. To answer the production responsiveness demands, the industry requires innovative solutions for highly adaptable and flexible production systems. This implies the need for reconfigurable/agile production systems, relying on, in the early production stage, fast ramp-up processes to reach full production capacity for new products and, in the mature production stage, an optimal operation execution at the plant level, while considering component-level degradation and required maintenance. To cope with such a challenge, introduction of advanced AI technologies in manufacturing plants is a must. Digitalization of factories is bringing several technological advances that facilitate the fulfilment of these requirements. To start with, there is an increased amount of sensor data (i.e. IIoT) and operation information that can now be deeply explored and exploited by AI systems: from reactive to proactive management. Furthermore, in order to embrace the upcoming Industry 5.0 it is necessary to enable effective ways for human intelligence to work in harmony with AI systems.
In line with these AI-based production challenges, the AI-PROFICIENT project which is a Research and Innovation Action (RIA), aspires to bring advanced AI technologies to manufacturing and process industry, while improving the production planning and execution, and facilitating the collaboration between humans and machines (Ethics by design approach). So, main objective is to bring the advanced AI technologies to manufacturing domain through an evolution from hierarchical and reactive decision making to self-learning and proactive control strategies, underpinned by predictive and prescriptive AI analytics at both component and system level, by cross-fertilizing edge and platform AI, while leveraging the human knowledge and feedback for reinforcement learning. This project belongs to the ICT-38 cluster (https://ai4manufacturing.com) grouping together all the projects accepted in this call.
The AI-challenges focused are refined in 3 more precise General Objectif (GOs) and 6 Scientific/Technical Objectives (STOs). These objectives served as a guideline for the project development by referring to Project Requirements, User Requirements, Functional Requirements, Functionalities and AI-services. This development is aligned with 8 WPs. In addition to demonstrate improved performance through AI-services and technologies, 3 different pilot sites have been chosen: Continental Sarreguemines (combiline machine), Ineos Geel and Cologne (production of polypropylene and polyethylene polymers). In relation to these three pilots and the WPs structuring, the project has been constructed on a mapping of the WPs to different Use Cases (UC) well representative of specific current production/maintenance/quality situations to be improved by AI. These improvements are more concretely related to 8 UCs: 5 for CONTI and 3 for INEOS (INEOS 1 and 2 at Geel Plant; INEOS3 at Cologne plant).
In line with these AI-based production challenges, the AI-PROFICIENT project which is a Research and Innovation Action (RIA), aspires to bring advanced AI technologies to manufacturing and process industry, while improving the production planning and execution, and facilitating the collaboration between humans and machines (Ethics by design approach). So, main objective is to bring the advanced AI technologies to manufacturing domain through an evolution from hierarchical and reactive decision making to self-learning and proactive control strategies, underpinned by predictive and prescriptive AI analytics at both component and system level, by cross-fertilizing edge and platform AI, while leveraging the human knowledge and feedback for reinforcement learning. This project belongs to the ICT-38 cluster (https://ai4manufacturing.com) grouping together all the projects accepted in this call.
The AI-challenges focused are refined in 3 more precise General Objectif (GOs) and 6 Scientific/Technical Objectives (STOs). These objectives served as a guideline for the project development by referring to Project Requirements, User Requirements, Functional Requirements, Functionalities and AI-services. This development is aligned with 8 WPs. In addition to demonstrate improved performance through AI-services and technologies, 3 different pilot sites have been chosen: Continental Sarreguemines (combiline machine), Ineos Geel and Cologne (production of polypropylene and polyethylene polymers). In relation to these three pilots and the WPs structuring, the project has been constructed on a mapping of the WPs to different Use Cases (UC) well representative of specific current production/maintenance/quality situations to be improved by AI. These improvements are more concretely related to 8 UCs: 5 for CONTI and 3 for INEOS (INEOS 1 and 2 at Geel Plant; INEOS3 at Cologne plant).
From both the Requirements and the UCs improvements expected, works done in WP1 allowed to clearly define the AI-Services/Technologies which have been developed in WP2, WP3, WP4 and supported in WP5 through the AI-PROFICIENT platform. These AI-Services are the following:
- S_DIA for “Diagnostic and anomaly detection service”
- S_HEA for “Health state evaluation service”
- S_PRO for “Component prognostic service”
- S_HYB for “Hybrid models of production processes and digital twin service”
- S_PRE for “Predictive production quality assurance service”
- S_ROO for “Root-cause identification service”
- S-EAR for “Early detection service”
- S_OPP for “Opportunistic maintenance decision-making service”
- S_GEN for “Generative holistic optimization service”
- S_LSL for “Future scenario based lifelong self-learning system service.”
- S_HUM for “Human feedback service”
- S_ETD for “Explainable and transparent decision-making service”.
These AI-Services have been investigated to be consistent with the modules already available in the AI4EU platform. They have been implemented (in link with the UCs) on two AI-PROFICENT platforms: one in Continental plant and the other in Geel plant. It means that AI-services have been implemented and used really in operational conditions (TR6-TRL7) by INEOS and CONTI in site. This is the major result of the project (referred to RIA), particularly with a view to the future exploitation of KERs. No platform has been set up on the Cologne site mainly due to industrial constraints. However, if some UCs have not been deployed on site, they have given rise to work and results that are more at the Research level than at the Innovation one. So, these UCs cannot be considered “as stopped” but “as not deployed on site” (off-line connections for some of them).
So, In relation to the UCs situation, at the end of the project, it was elaborated 9 KERs issued of the AI-Services developed in the WP2, WP3, WP4 then implemented in the AI-PROFICIENT platforms and finally evaluated in WP6. In that way, all the WPs have been achieved in consistence with the work planned. More precisely with regards to WP7 dedicated to dissemination and communication activities, most of the KPIs targeted for each of these activities have been achieved.
- S_DIA for “Diagnostic and anomaly detection service”
- S_HEA for “Health state evaluation service”
- S_PRO for “Component prognostic service”
- S_HYB for “Hybrid models of production processes and digital twin service”
- S_PRE for “Predictive production quality assurance service”
- S_ROO for “Root-cause identification service”
- S-EAR for “Early detection service”
- S_OPP for “Opportunistic maintenance decision-making service”
- S_GEN for “Generative holistic optimization service”
- S_LSL for “Future scenario based lifelong self-learning system service.”
- S_HUM for “Human feedback service”
- S_ETD for “Explainable and transparent decision-making service”.
These AI-Services have been investigated to be consistent with the modules already available in the AI4EU platform. They have been implemented (in link with the UCs) on two AI-PROFICENT platforms: one in Continental plant and the other in Geel plant. It means that AI-services have been implemented and used really in operational conditions (TR6-TRL7) by INEOS and CONTI in site. This is the major result of the project (referred to RIA), particularly with a view to the future exploitation of KERs. No platform has been set up on the Cologne site mainly due to industrial constraints. However, if some UCs have not been deployed on site, they have given rise to work and results that are more at the Research level than at the Innovation one. So, these UCs cannot be considered “as stopped” but “as not deployed on site” (off-line connections for some of them).
So, In relation to the UCs situation, at the end of the project, it was elaborated 9 KERs issued of the AI-Services developed in the WP2, WP3, WP4 then implemented in the AI-PROFICIENT platforms and finally evaluated in WP6. In that way, all the WPs have been achieved in consistence with the work planned. More precisely with regards to WP7 dedicated to dissemination and communication activities, most of the KPIs targeted for each of these activities have been achieved.
The progress beyond the state of the art is mainly visible through the innovations supported with the 9 KERs (e.g. exploited as open-source software or as add-on functionalities to existing commercial systems; Integrated system) created:
1- Surrogate data driven model based on explainable AI,
2- Connected worker Additive Check application using innovative human-machine interfaces,
3- Data quality analysis module: GO-QData
4- Natural Human-Computer Interaction
5- Process anomaly detection: GO-QNormality
6- Multi-objective optimizer generative and MI(N)LP optimization solvers
7- Semantic Accountability Tool
8- Data-driven predictive AI analytics for prognostics based on DL
9- Machine vision for dimensional measurement and positioning with small “bad” datasets
As several experimentations and assessments of the AI-services/technologies leading to these KERs have been done on two sites in operational conditions and with regards to UC concerned, the project is able to demonstrate:
- General positive impacts of these AI-solutions on Products and Service, Humans role.
- Specific positive impacts in link for example with OOE increasing, extending operation life time
- User experience interest in line with these solutions,
- Ethical recommendations to be used for other industrial context. "Ethics by design" approach is really an innovation that the project has to promote.
- Best practices and lessons learned.
1- Surrogate data driven model based on explainable AI,
2- Connected worker Additive Check application using innovative human-machine interfaces,
3- Data quality analysis module: GO-QData
4- Natural Human-Computer Interaction
5- Process anomaly detection: GO-QNormality
6- Multi-objective optimizer generative and MI(N)LP optimization solvers
7- Semantic Accountability Tool
8- Data-driven predictive AI analytics for prognostics based on DL
9- Machine vision for dimensional measurement and positioning with small “bad” datasets
As several experimentations and assessments of the AI-services/technologies leading to these KERs have been done on two sites in operational conditions and with regards to UC concerned, the project is able to demonstrate:
- General positive impacts of these AI-solutions on Products and Service, Humans role.
- Specific positive impacts in link for example with OOE increasing, extending operation life time
- User experience interest in line with these solutions,
- Ethical recommendations to be used for other industrial context. "Ethics by design" approach is really an innovation that the project has to promote.
- Best practices and lessons learned.