DEPOSIA focuses on the detection and geolocation of various radio frequency signal sources in order to thwart attacks on connected systems and infrastructures. The sources considered are elements which by their characteristics or their position, present an illicit character and which threaten the people security or the infrastructures. For outdoor cases, we consider drones flying over forbidden areas, telecommunication jammers, spoofing signal transmitters or wireless connected sensors used to introduce false data in monitoring platforms. For indoor cases, we also consider jamming or spoofing sources that can cause denial of service within networks or infrastructures, or fake access points that aim to carry out man-in-the-middle attacks to intercept information. In this proposal, the indoor and outdoor use cases are considered separately in order to design monitoring infrastructures adapted to each case. For the outdoor case, we consider a surveillance architecture that could join the already existing cellular or WLAN communication infrastructures. In particular, with 5G technology and the higher employed frequencies, cellular networks are evolving towards finer meshes and have interfaces with the core network at each of their nodes. Thus, these interface points, equipped with receivers dedicated to monitoring, could enable the routing of monitoring data to centralized platforms, feeding an Artificial Intelligence for analysis, anomaly detection and source geolocation. For the indoor case, we consider a distributed monitoring architecture deployed within a building, based on SDR sensors and a data centralization and synchronization network. In these two cases, we envisage an Artificial Intelligence working on data evolving in three dimensions : time, space and direction, all for data of different natures, namely those from the physical layer and the data link layer. Whether for indoor or outdoor configurations, the algorithms that will constitute the Artificial Intelligence will be based on learning approaches that will correspond to Machine Learning and Deep Learning algorithms. These algorithms will deal with the problems of detecting attacks and locating illicit sources. These algorithms will have to take into account: the evolutionary aspect brought by the non-fixed character in time of the attacks and the non-fixed location aspect of the localization of the source of the attack. A first Artificial Intelligence will be dedicated to data analysis and anomaly detection, i.e., highlighting the suspicious nature of the data, and a second Artificial Intelligence will be dedicated to extracting the location information of the attack source. Due to the multi-layered nature of the data, model aggregation algorithms will be deployed in order to homogenize the decision process.

The XPM project aims to integrate explanations into Artificial Intelligence (AI) solutions within the area of Predictive Maintenance (PM). Real-world applications of PM are increasingly complex, with intricate interactions of many components. AI solutions are a very popular technique in this domain, and especially the black-box models based on deep learning approaches are showing very promising results in terms of predictive accuracy and capability of modelling complex systems. However, the decisions made by these black-box models are often difficult for human experts to understand – and therefore to act upon. The complete repair plan and maintenance actions that must be performed based on the detected symptoms of damage and wear often require complex reasoning and planning processes, involving many actors and balancing different priorities. It is not realistic to expect this complete solution to be created automatically – there is too much context that needs to be taken into account. Therefore, operators, technicians and managers require insights to understand what is happening, why it is happening, and how to react. Today’s mostly black-box AI does not provide these insights, nor does it support experts in making maintenance decisions based on the deviations it detects. The effectiveness of the PM system depends much less on the accuracy of the alarms the AI raises than on the relevancy of the actions operators perform based on these alarms. In the XPM project, we will develop several different types of explanations (anything from visual analytics through prototypical examples to deductive argumentative systems) and demonstrate their usefulness in four selected case studies: electric vehicles, metro trains, steel plant and wind farms. In each of them, we will demonstrate how the right explanations of decisions made by AI systems lead to better results across several dimensions, including identifying the component or part of the process where the problem has occurred; understanding the severity and future consequences of detected deviations; choosing the optimal repair and maintenance plan from several alternatives created based on different priorities, and understanding the reasons why the problem has occurred in the first place as a way to improve system design for the future.

The APPRENTIS project concerns the safety of industrial or port areas presenting risks (e.g., fire, explosion or toxic leakage). The operational objective is to provide a decision support software tool to plan monitoring and rescue patrols. This tool will minimize the cost of the patrols carried out by mobile agents (as drones or automated vehicles) by optimizing physical and financial resources based on the analysis of data flows. The questions we would like to answer are as follows: • During monitoring, how many mobile agents are required to perform a given set of measurements at given positions? What sensors should each of these agents equip? How to define the patrols of each of the agents in order to meet the overall monitoring requirements? • In the event of an incident, how to use these same monitoring agents to quickly obtain relevant information on the incident, the damage and any victims? How to transport and distribute rescue supplies with the help of intervention agents? Finally, how can we jointly and effectively use monitoring and intervention agents? The originality of the method proposed to solve these problems is based on modeling aspects and a resolution methodology that are derived from discrete event systems (DESs) and artificial intelligence (AI). This dual approach is motivated by the exponential complexity of the problem which appears when the problems of configuration and planning of the patrols of each of the agents are combined, the latter depending on the evaluated configuration. The expected result of the APPRENTIS project is a demonstrator that can serve as a basis for the development of a software devoted to the configuration of the monitoring and intervention patrols from a catalog of equipment, the description of the infrastructure, and the patrol specifications. We target, in particular, 3 types of audiences: 1) Companies with SEVESO classified sites (156 sites in Hauts de France region, 86 sites in Normandy region, 99 sites in the PACA region and 94 sites in the Ile de France region) which are called upon to strengthen the monitoring of their installations; 2) Organizations and local authorities in charge of crisis intervention (SDIS, urban communities, associations as ORMES); 3) Economic interest groups which are concerned with the production, transport or storage of products at risk (Ports of Le Havre, Rouen and Paris - HAROPA, Grand Port Maritime de Marseille - GPMM). The consortium of partners (ULHN - GREAH - EA 3220, AMU - LIS - UMR 7020, IMTLD, USPN - LURPA - EA 1385) was formed on the basis of the partners’ experience in the risk management and in the implementation and use of DES and AI tools. ULHN in Normandy region and IMTLD in Hauts de France region are located in the two regions targeted by the call RA-SIOMRI. AMU and USPN are located in two large and densely populated cities for which the potential impacts of industrial incidents are particularly serious. Finally, the city of Marseille offers similarities with the city of Le Havre through its port activity, an additional vector of risks due to the storage of hazardous materials, and through its concentration of SEVESO industrial sites near residential areas (Fos-sur-Mer near Marseille and Tancarville near Le Havre). The longer-term challenge we initiate here is to coordinate the means of monitoring and intervention in an automated way by combining predictive and decision-making models, and using model-based methods as well as database-based methods.

Agriculture faces a triple challenge: ensuring food production for the world's growing population, ensuring economic and environmental sustainability of farming practices, and making up for the lack of labor for hard, dangerous, and repetitive work. By ensuring precision and availability, agricultural robotics brings environment and productivity closer together. However, scientific and technological developments related to agricultural robotics have mainly focused on a single robot. Very little work has been done on distributed, scalable and autonomous coordination and route guidance of a fleet of heterogeneous agricultural robots without a centralized coordination center while allowing for seamless dynamic change of the structure of the fleet and distributed and scalable adaptation in real time to the dynamically changing environment. The AGRIFLEETS project aims to fill this void by proposing to develop a distributed multi-agent architecture for the coordination of a fleet of mobile agricultural robots. To this end, we will focus on dynamic, scalable and distributed multi-agent solution methods for multi-index task assignment and the dynamic vehicle routing problem. This approach will allow us to achieve autonomous real-time coordination of fleets of agricultural mobile robots and improve fleet performance and efficiency while minimizing dependence on human labor.