Use Cases
The SPACE Business Unit of CS GROUP brings its technical expertise to design & develop the use case “Improvement of flash flood forecasting thanks to the use of AI”. This use case is about testing the feasibility of modelling flood events by neural network, and to evaluate the sensitivity of the model to the available data.
For this purpose, we propose the following 4-step approach:
- Establish a resulting fine-scale hydrodynamic reference model on the selected area with historical episodes.
- Develop and adjust the AI model with a minimum number of parameters, easily accessible.
- Qualify the AI model by comparison with theresults of the hydrodynamic model, considered as the reference to be reached.
- Qualify the AI model by testing the impact and sensitivity to input parameters, weights, etc. to avoid over-sampling, and/or to avoid missing some key parameters.
For its top-notch representativeness of flash flood events in urban areas, the city of Nîmes was chosen. The existence of the ESPADA system, the network of sensors in place, the richness of historical data makes it an excellent ground for the development of the model.
If the results are up to expectations, the methodology could be applied to other urbanized areas, not benefiting from these input data.
It is also expected that the CS METIS platform will be used for the implementation and configuration of the hydrodynamic reference models as well as the integration of the AI models in order to optimize them by reducing the input parameters, the number of announcement types, while taking advantage of the ExtremeXP environment (testing and validation of the use case on the external platform that will be provided by a project partner).
Attackers evolve perpetually by utilizing advanced techniques and refining of implements and tactics for performing targeted penetration strategies directed towards internal and external access points of organizations. The main challenge exists in having reliable and robust detection i.e., high accuracy and low false positives. It is necessary to develop methods that can utilize the correlation between suspicious information flows that arise during an attacker campaign and break their kill chain. The main challenge remains is the semantic gap between the low-level alerts and high-level kill chain view of the attacker’s behaviour. In our analysis, we aim to bridge the semantic gap with the behavioural patterns in MITRE ATT&CK framework such that we can effectively match the low-level alerts to the TTPs, discover the correlation between the attack steps and reduce the false positives. In this use case, we propose an early multistage real-time attack detection with an awareness of adversary Tactics, Techniques, and Procedures (TTP) as suggested in the below architecture.
The first use case is the alert detection. The scenario starts with the reception of sensors videos and the analysis of the streams.
If a fire is detected by an algorithm (Yolo for example), an alert is sent to the operator with the location and a short video containing the fire.
The operator will check the alert and send a reply to the system.
- If the alert is a wrong one, the operator replies to few questions to identify the non-validation.
- If the alert is valid, the operator will ask to the closest on-field user to go to the alert location to confirm the alert.
The variability points are: the video characteristics (data rate, frames per seconds and definition) , the fire detection algorithm (frames par second analysed) and the user routing algorithm (user status, user location, type of vehicle…)
The second use case receive the alert from the first one. Depending on the type of alert, the system will propose to the operator a list of available and closest users to the alert location.
The operator will confirm, modify or refuse the proposition.
When the list is validated, the system will ask directly to the users their availability and select an other user if the answer is negative. In case of availability, the selected user will receive all data necessary to the mission.
The variability points are the routing algorithm and the parameters for the routing selection (availability, location, type of user…)
Moby’s use case, UC4, focuses on enhancing travel pattern prediction accuracy to support transportation authorities, planners, and policymakers by providing actionable insights for optimizing mobility services and policy decisions through flexible, data-driven modeling.
The objectives include streamlining model selection with user-driven AutoML, integrating complex analytics workflows for heterogeneous environments by combining various data sources, and delivering custom policy analysis through adaptive analytics. The use case leverages a dataset of millions of location points, along with sensor data labelling to enrich predictions and employs experimentation-driven analytics to refine models based on user input.
The Use Case aims to reduce manual travel diary entries by user, through improving the ML modules by assessing the efficiency of different versions of the algorithm and different workflows within the experimentation engine of ExtremeXP, aiming to increase algorithm efficiency. In addition, data visualization modules will be integrated within the workflow to enable the user-in-the-loop concept and improve customer satisfaction.
Maximizing machine uptime in an industrial plant is crucial for attaining high overall system availability and maintaining competitiveness. The Industry 4.0 paradigm has introduced novel techniques and methods to achieve this objective, including predictive maintenance, remaining lifetime assessment, and critical machine element diagnosis.
The key is to have precise information and avoid erroneous warnings that lead to a decrease of the production capacity. With this in mind, the objective is to have a dashboard for the technical service team that allows them to take decisions over incipient and imminent failures of the critical components of the machine, such as heads, balls screws or axis.
