Periodic Reporting for period 2 - STAR (Safe and Trusted Human Centric Artificial Intelligence in Future Manufacturing Lines)
Reporting period: 2022-07-01 to 2023-12-31
STAR was a joint effort of AI and digital manufacturing experts towards enabling the deployment of standard-based secure, safe reliable and trusted human centric AI systems in manufacturing environments and researches and integrates leading edge AI technologies with wide applicability in manufacturing environments, including:
• Active learning (AL) systems that boost safety and accelerate the acquisition of knowledge.
• Simulated reality systems that accelerate Reinforcement Learning (RL) in human robot collaboration scenarios.
• Explainable AI (XAI) systems that boost the transparency of industrial systems and increase trust in them.
• Human Centric digital twins enabling worker monitoring for safer and trustful production processes.
• Advanced RL techniques for optimal navigation of mobile robots and for the detection of safety zones in industrial plants.
• Cyber-defence mechanisms for sophisticated poisoning and evasion attacks against deep neural networks operating over industrial data.
STAR’s focus on these research areas placed the project at the forefront of the global research in AI in general and in digital manufacturing in particular. STAR became a catalyst for the deployment of the most advanced AI systems in real-life manufacturing environments, through researching and providing effective ways for moving novel research results from the partners’ labs to the shop-floor. The project leveraged background projects and results of the partners in the above areas, which ensured research excellence and enabled STAR to stand out from similar research initiatives worldwide.
STAR had set 8 overall objectives. These were:
O1: Reference architecture and platform implementation for safe, reliable, secure and human centric AI in manufacturing
O2: Transparent & XAI in manufacturing
O3: Reliable AI and human-centric knowledge acquisition based on simulated reality and active learning
O4: Human centred simulations and digital twins for safe AI systems in manufacturing
O5: Cyber security and data reliability for AI systems in manufacturing environments
O6: Real-life integration, validation and evaluation in production lines
O7: Legal, regulatory and policy making guidelines for ethical AI
O8: Market platform and digital innovation hub establishment - integration with AI4EU
All these objectives have been achieved by the end of the project.
• Active learning (AL) systems that boost safety and accelerate the acquisition of knowledge.
• Simulated reality systems that accelerate Reinforcement Learning (RL) in human robot collaboration scenarios.
• Explainable AI (XAI) systems that boost the transparency of industrial systems and increase trust in them.
• Human Centric digital twins enabling worker monitoring for safer and trustful production processes.
• Advanced RL techniques for optimal navigation of mobile robots and for the detection of safety zones in industrial plants.
• Cyber-defence mechanisms for sophisticated poisoning and evasion attacks against deep neural networks operating over industrial data.
STAR’s focus on these research areas placed the project at the forefront of the global research in AI in general and in digital manufacturing in particular. STAR became a catalyst for the deployment of the most advanced AI systems in real-life manufacturing environments, through researching and providing effective ways for moving novel research results from the partners’ labs to the shop-floor. The project leveraged background projects and results of the partners in the above areas, which ensured research excellence and enabled STAR to stand out from similar research initiatives worldwide.
STAR had set 8 overall objectives. These were:
O1: Reference architecture and platform implementation for safe, reliable, secure and human centric AI in manufacturing
O2: Transparent & XAI in manufacturing
O3: Reliable AI and human-centric knowledge acquisition based on simulated reality and active learning
O4: Human centred simulations and digital twins for safe AI systems in manufacturing
O5: Cyber security and data reliability for AI systems in manufacturing environments
O6: Real-life integration, validation and evaluation in production lines
O7: Legal, regulatory and policy making guidelines for ethical AI
O8: Market platform and digital innovation hub establishment - integration with AI4EU
All these objectives have been achieved by the end of the project.
In respect to the set of eight project objectives that have been defined all have been achieved.
O1: The final version of the reference architecture (STAR Reference Architecture and Blueprints-Initial version) includes specifics on implementation, deployment, and a wider range of blueprints for trusted AI systems in industrial settings.
O2: Experiments and validation were performed on XAI techniques. The final design of the XAI component was completed, aligning with use cases and available datasets. Solutions were developed for XAI tasks related to visual inspection needs, with methods adapted for the use cases. Trust evaluation of users in the Human-in-the-AI-Loop was conducted, alongside an innovative XAI solution for timeseries classification and research showing the robustness of LIME-masked images against evasion attacks.
O3: Simulated Reality approaches were enhanced to address data scarcity and increase robustness in visual perception for AI systems. Active Learning approaches were integrated for user interaction strategies, with experiments on AL and XAI for quality inspection support. Various user interaction methods were tested to combat worker fatigue.
O4: STAR focused on developing solutions for AI-based human-in-the-loop production processes. Initial efforts established foundational elements, leading to the architecture and prototype of the Human Digital Twin (HDT). Progress includes advancing AI modules for worker simulation, monitoring, decision-making support, and process optimization. Demonstrators have been developed and integrated into the HDT Core Infrastructure, showcasing collaborative advancements in AI technologies within the project.
O5: STAR also focused on developing the AI cyber-defence components to safeguard against AI attacks. An experimental testbed to assess adversarial methodologies, highlighting the impact of attacks on AI/DNN setups was created whereas a decentralised reliability component was defined and a blockchain infrastructure for secure data sharing in smart manufacturing was developed. The STAR Security Policy Manager using OPA and enhanced the Risk Assessment and Mitigation Engine was finalised.
O6: Advancement in STAR technologies were integrated and deployed across the three pilots. Over eight Use Cases were developed and implemented.
O7: STAR provided an extensive analysis of over 35 standards, regulations, and directives in Privacy, Safety, and Cybersecurity categories. STAR partners have produced more than 40 publications (in total) in the scientific journals and conferences. At the same time STAR has produced one Open Access Book and has led the production and contributed with chapters in the joint ICT-38-2020 projects’ Open access Book. Furthermore, a practical AI system trustworthiness auditing framework "scorecard" for manufacturers was developed in addition to the creation of several courses.
O8: The implementation of the market platform was finalised. Project assets and Pilot Use Cases are included in the "Success Stories" section. The marketplace integrates tools and services like the Multimodal Workers’ Training Platform and the courses that were prepared during the project. At the same time it provides external content and links to learning platforms, communities, and manufacturing ecosystems.
O1: The final version of the reference architecture (STAR Reference Architecture and Blueprints-Initial version) includes specifics on implementation, deployment, and a wider range of blueprints for trusted AI systems in industrial settings.
O2: Experiments and validation were performed on XAI techniques. The final design of the XAI component was completed, aligning with use cases and available datasets. Solutions were developed for XAI tasks related to visual inspection needs, with methods adapted for the use cases. Trust evaluation of users in the Human-in-the-AI-Loop was conducted, alongside an innovative XAI solution for timeseries classification and research showing the robustness of LIME-masked images against evasion attacks.
O3: Simulated Reality approaches were enhanced to address data scarcity and increase robustness in visual perception for AI systems. Active Learning approaches were integrated for user interaction strategies, with experiments on AL and XAI for quality inspection support. Various user interaction methods were tested to combat worker fatigue.
O4: STAR focused on developing solutions for AI-based human-in-the-loop production processes. Initial efforts established foundational elements, leading to the architecture and prototype of the Human Digital Twin (HDT). Progress includes advancing AI modules for worker simulation, monitoring, decision-making support, and process optimization. Demonstrators have been developed and integrated into the HDT Core Infrastructure, showcasing collaborative advancements in AI technologies within the project.
O5: STAR also focused on developing the AI cyber-defence components to safeguard against AI attacks. An experimental testbed to assess adversarial methodologies, highlighting the impact of attacks on AI/DNN setups was created whereas a decentralised reliability component was defined and a blockchain infrastructure for secure data sharing in smart manufacturing was developed. The STAR Security Policy Manager using OPA and enhanced the Risk Assessment and Mitigation Engine was finalised.
O6: Advancement in STAR technologies were integrated and deployed across the three pilots. Over eight Use Cases were developed and implemented.
O7: STAR provided an extensive analysis of over 35 standards, regulations, and directives in Privacy, Safety, and Cybersecurity categories. STAR partners have produced more than 40 publications (in total) in the scientific journals and conferences. At the same time STAR has produced one Open Access Book and has led the production and contributed with chapters in the joint ICT-38-2020 projects’ Open access Book. Furthermore, a practical AI system trustworthiness auditing framework "scorecard" for manufacturers was developed in addition to the creation of several courses.
O8: The implementation of the market platform was finalised. Project assets and Pilot Use Cases are included in the "Success Stories" section. The marketplace integrates tools and services like the Multimodal Workers’ Training Platform and the courses that were prepared during the project. At the same time it provides external content and links to learning platforms, communities, and manufacturing ecosystems.
STAR aimed to advance AI in manufacturing through cutting-edge AI systems implementation and customization for popular applications, integrating advanced AI techniques for transparency and robustness, deploying advanced RL systems for human-AI interaction, and incorporating novel XAI systems for transparency and trustworthiness.
The social impacts foreseen from STAR are summarised as following:
1) Ethical AI in Manufacturing: STAR boosts the development, deployment and operation of Ethical AI systems for smart manufacturing based on solutions that boost the security, the trustworthiness, the robustness and the safety of AI systems in digital manufacturing.
2) Worker Safety: STAR has a positive impact on the safety of factory workers, through providing the means for avoiding hazardous situations that may result from malfunctions or poor operation of robots and other AI systems, especially in cases where humans collaborate with them.
3) Increased Workers’ and Manufacturers’ Trust in AI systems: STAR boosted the robustness, safety and trustworthiness of AI systems in manufacturing environments, which increasse stakeholders’ trust in AI systems. This could greatly facilitate the wider deployment and use of AI in manufacturing.
4) Compliance to Safety Standards and Regulations: STAR discusses from a legal and ethical viewpoint the design and use of products made with Artificial Intelligence (AI) technology within the context of smart manufacturing and facilitates manufacturers and providers of AI solutions in complying with safety regulations and standards, which will help stakeholders create safe and compliant environments, while reducing the probability of adverse effects of non-compliance (e.g. regulatory penalties).
The social impacts foreseen from STAR are summarised as following:
1) Ethical AI in Manufacturing: STAR boosts the development, deployment and operation of Ethical AI systems for smart manufacturing based on solutions that boost the security, the trustworthiness, the robustness and the safety of AI systems in digital manufacturing.
2) Worker Safety: STAR has a positive impact on the safety of factory workers, through providing the means for avoiding hazardous situations that may result from malfunctions or poor operation of robots and other AI systems, especially in cases where humans collaborate with them.
3) Increased Workers’ and Manufacturers’ Trust in AI systems: STAR boosted the robustness, safety and trustworthiness of AI systems in manufacturing environments, which increasse stakeholders’ trust in AI systems. This could greatly facilitate the wider deployment and use of AI in manufacturing.
4) Compliance to Safety Standards and Regulations: STAR discusses from a legal and ethical viewpoint the design and use of products made with Artificial Intelligence (AI) technology within the context of smart manufacturing and facilitates manufacturers and providers of AI solutions in complying with safety regulations and standards, which will help stakeholders create safe and compliant environments, while reducing the probability of adverse effects of non-compliance (e.g. regulatory penalties).