Project start | 1^st January 2021

PI: Dr. Maria Chiara Leva

The European Commission’s guidelines on ethics in artificial intelligence (AI), published in April 2019, recognised the importance of a ‘human-centric’ approach to AI that is respectful of European values. Dedicated training schemes to prepare for the integration of “human-centric” AI into European innovation and industry are now needed. AIs should be able to collaborate with (rather than replace) humans. Safety critical applications of AI technology are “human- in-the-loop” scenarios, where AI and humans work together, as manufacturing processes, IoT systems, and critical infrastructures.

The concept of Collaborative Intelligence is essential in these scenarios. The CISC EID will nurture and train 14 world class-leading Collaborative Intelligence Scientists for safety critical situations and provide a blue-print for postgraduate training in this area. The development of Collaborative Intelligence systems requires an interdisciplinary skillset blending expertise across AI, Human Factors, Neuroergonomics and System Safety Engineering. This inter-disciplinary skill- set is not catered for in traditional training courses at any level.

CISC is a Marie Curie Training Network funded by the European Commission to hire and train Early Stage Researchers (ESR) or PhD student as Collaborative Intelligence Scientists with the expertise and skillset necessary to carry-out the major tasks required to develop a Collaborative Intelligence system.

The CISC training programme will develop Collaborative Intelligence Scientists with the scientific expertise and complementary so-called “soft” skillset necessary to address this need:

     1. Using data analytics and AI to create novel human-in-the-loop automation paradigms to support decision making and or anticipate critical scenarios (Theme 1);

     2. Designing and implementing processes capable of monitoring interactions between automated systems and the humans destined to use them (Theme 2: Human Factors/Neuroergonomics);

     3. Modelling the dynamics of system behaviours for the manufacturing processes, IoT sensor systems, and critical infrastructures (Theme 3: System Safety Engineering);

     4. Managing the Legal and Ethical Implications of AI algorithms, and the use of physiologyrecording wearable sensors and human performance data in them (Theme 4);

The accelerating progress in AI and automation is bringing further opportunities for users, businesses, and economy. Europe, while making progress, is behind United States and China, therefore it is now imperative for the EU to develop the human capital needed to cope with its 'human-centric' approach to AI and to face the multi-facets impacts of this choice. This relentless march of technology is opening new possibilities for AI driven automation in domains as diverse as manufacturing, process industry, healthcare and transport. While all these sectors anticipate benefits, in terms of cost, productivity and safety, few understand the importance of fully considering how to interface AIs with the humans that are supposed to use them in order to realise the anticipated benefits, and even fewer know how to address these new types of human-machine collaboration. CISC will train researchers capable of addressing this shortcoming and contribute to the EU human-centric approach to artificial intelligence by providing two key profiles:

1. An AI analyst with training in Human Factors/Neuroergonomics and system safety engineering for automation. (Profile shaped for companies such as the one represented by IMR, DIGITALSME, MATHEMA, HUGIN SCCH, IVECO etc. around Theme 1,2 predominantly & theme 3 secondarily).
2. A System Engineer with human factor background that will be trained in data analytics, and AI algorithms applied in automation. (Profile shaped for companies such as PILZ, IVECO, Adient, IMR, YOKOGAWA, EPRI., etc. around Theme 3,2 predominantly & theme 1 secondarily).

More information about the project: CISC project

Human-AI Teaming Platform for Maintaining and Evolving AI Systems in Manufacturing

Project Start | 1^st January 2021

PI: Dr. Maria Chiara Leva

Smart Manufacturing is believed to play a critical role in maintaining the competitiveness of organisations, by supporting them at different levels such as process optimisation, resource efficiency, predictive maintenance and quality control. Nevertheless, AI technologies which are currently and rapidly penetrating industrial sectors at those levels remain essentially narrow AI systems. This is due to the lack of self-adaptiveness in the AIs capability to assimilate and interpret new information outside of its predefined programmed parameters. This means that AI systems are tailored for solving specific tasks in a specific predefined setting and changes in the underlying setting usually require system adaption ranging from fine-grained parameter adaptations to fully-fledged re-design and re-development of AI systems.

TEAMING_AI project aims at a human AI teaming framework that integrates the strengths of both, the flexibility of human intelligence and scale-up capability of machine intelligence. Human AI teaming is equally motivated to meet the increased need for flexibility in the maintenance and further evolution of AI systems, driven by the increasing personalization of products and services, as well as tackling the barriers of user acceptance and ethical challenges involved in the collaborative environments where artificial intelligence will be used, in order AI can be considered as 'teammate' rather than as a threat.

Teaming AI means collaborative cooperation between humans and the AI system, in that sense; we cannot consider the human as a passive element in the system, but as an element that has an active role (human in the loop) providing essential information for improving the capabilities of the IA system and for enhancing the Situational Awareness of the team.

Members of HFISS will support the 'human-centric' approach to AI in Teaming.AI on USE case 1 and use case 2 by informing the mapping of the tacit knowledge and guiding the necessary step to adopt a human-in-the-loop approach for the AI algorithm to be used in training the machine deployed in use case 1 and 2. It will develop the process necessary in involving the humans and supporting the development of an explanation-based collateral systems to explain the behaviour of the AI algorithms used in terms that humans can understand--from how they interpreted their input to why they recommended a particular output.

In short, the expected outcomes for the High-precision manufacturing of automotive parts (UC1 and UC2) are:

• To increase Situational awareness by actively informing the machine status to the operator. ​
• To increase teaming by involving the operator for verification.
• To create shift hand-over protocol showing machine status at all time​
• To create an interface for incident monitoring.

In the case of Use Case 3 related to high-precision manufacturing of large parts:​

• To improve interplay between AI-controlled machine tasks and human labour.
• Ergonomic risk assessment of the two simultaneous tasks in terms of static loads and repetitive strains as well as execution time​ taking into account:
  • Research Key Aspect: Workflow level, team and task scheduling and synchronization

The TEAMING.AI project will be run over 36 months with a work plan divided into 9 Work Packages. Work Packages from 1 to 5 are devoted to the development of new technology to enhance the interaction between human and machine. Furthermore, Work Packages 6 and 7 wrap the development of 3 use case scenarios. Finally, two final Work Packages (8 and 9) will work respectively on the dissemination, exploitation of results and coordination of the project in a transversal way.

More information about the project: TEAMING.AI

This is an ERASMUS + SAFETY4VET project that runs Up until July 2024. The contribution of HFISS focuses on the human factors and Ergonomics with researchers working on training needs analysis for the design of this vocational training program.

Safety of Machinary approach

By Mario Di Nardo

Each single event or sequence of unforeseen events can cause damage to people, material assets or the surrounding environment: it follows that, whatever the context to which it refers, the risk represents a component that is always present. Consequently, it is necessary, at the organizational level:

1. identify risk factors through systemic procedures;
2. evaluate them through quantitative methods;
3. plan and implement preventive and protective measures to ensure the safety of people, property and the environment, as well as avoid economic and productive losses if an unwanted event occurs.

The course will briefly introduce ISO 31000 and focus on ISO 31010, which discusses risk assessment techniques. In this context, the industry is a Compex system where the interaction between heterogeneous factors (dangerous substances, human factors, managerial and organizational aspects) can give rise to process deviations that can result in failures if not properly managed. Therefore, it is advisable to identify how the aforementioned deviations can occur so the system can survive. Problems and complexes will be analyzed using the Sytem Dynamics analysis model to provide a possible interpretation. Finally, a quantitative  case study will be proposed as an application.

Title: Developing a real-time mental workload assessment method of Air Traffic Controllers based on behavioral measures.

Project ID: EPSPD/2022/151

PI: Dr. Enrique Muñoz de Escalona Fernandez

Air Traffic Controllers’ (ATCo) Mental Workload (MW) is likely to remain the single greatest functional limitation on the capacity of the Air Traffic Management (ATM) System. MW in the ATM domain has been attempted to be estimated and monitored using subjective, physiological and behavioral measures. However there is currently no accepted single method deployed in the industry to assess and monitor MW, fatigue and the effect it has on performance, even if the industry has now issued a requirement for active fatigue risk management processes as suggested by the International Civil Aviation Organization (ICAO). The disadvantages highlighted within the State-of-the-art for subjective and physiological measures is related to how obtrusive and impractical they can be to use in real work scenarios. First, they both interfere with ATCos’ performance: 1) online measuring of MW require attentional resources to be focused on introspection every time subjective reports of MW are requested and 2) most physiological measures need to be collected by using intrusive equipment which would ultimately interfere with task development and even with experienced MW. Secondly, one outstanding feature of subjective measures is that they may be distorted. For these reasons, the industry and scientific community need to develop a MW calculation model that can be based on an assessment of ATCos’ recordable behavioral measures (considering ATCos’ communications patterns and their interactions with the ATM systems) that can be deployed unobtrusively in an ecologically valid environment.  The main advantage of this model is to overcome current limitations primarily because communication patterns and interactions with the ATM systems can be analyzed indirectly (and in real-time) though the logs of the ATM system automation, with equipment that is already an integrated tool of ATCos’ tasks; in addition those behavioral measures cannot be distorted. 

Hence, the aims of the project are all to be considered interim scientific breakthrough consisting of: 

1. The development of a computational model of ATCo’s MW based on the behavioural data recordable through the ATM automation, about 1) ATCos’ communication patterns and 2) their recorded clearance actions and choices. 
2. Ensuring the model can be used to unobtrusively calculate real-time ATCos’ MW. 
3. Validating the model using well-stablished physiological measurements based on voice and eye-tracking parameters, alongside subjective MW reports. 

This will enable the industry to develop and test supporting mechanisms for task complexity variations mitigating the disastrous effect of drops in human performance. 

Title: Resilience assessment of pilot Hydrogen facility for production, storage and distribution to support safer green energy transition in Ireland.

Project ID: EPSPD/2023/314

PI: Dr. Hector Diego Estrada-Lugo

Traditionally, the design of critical infrastructure systems has been based on an assessment of the risks that potential hazards may pose to such systems. This approach focuses, therefore, on improving the recovery of the existing systems through incremental changes to fix observed failures after analysing the root causes. A resilience-by-design framework for updating the design methods of critical infrastructure to anticipate future scenarios. The approach relies on identifying solutions to potential failures in the system to enhance persistence, adaptation, and potential transformations in the design stage. The proposed framework offers a system-of-systems perspective of resilience that consists of the modelling and quantification of the components and sub-components that may affect the performance of the critical infrastructure.

According to the integrated framework for resilience quantification proposed by Estrada-Lugo, et.al. (2020) and Santhosh&Patelli(2021), every infrastructure system has its own performance metrics such as reliability, availability, maintainability, and safety. For a continuously operating systems or infrastructure, the widely accepted measure of performance is availability over reliability as the system is usually restored after a disruptive event. Information regarding the reliability and safety of hydrogen facilities can be adopted for determining the risk of dangerous substances such as hydrogen because it considers and quantifies both the realistic hazards and the associated system conditions and failures. A number of hydrogen system safety data sources can be consulted (e.g., international standards, H2Tools, Hydrogen Incidents and Accidents Database (HIAD), among others) to quantify the hazards related to these installations.

The proposed resilience framework of Estrada-Lugo and Santhosh, although designed for a wide spectrum of complex engineering systems, e.g., nuclear power plants, has not been applied to hydrogen related facilities. Further to the resilience quantification method, improvements can be done to the analysis by including costs of resilience. Such costs are related to the implementation costs of recovery actions to bring back the system to operational performance from a disrupted state after the occurrence of a disruptive event.