The Data Science for Industry and Physics (DSIP) Research Unit focuses on applying Data Science methodologies and approaches to develop predictive machine-learning models from heterogeneous data. DSIP is actively collaborating with industrial partners and research organizations. DSIP is involved in developing Deep Learning solutions for time-series analysis and forecasting in the context of condition monitoring and predictive maintenance for industrial process forecasting and control. In the earth and climate sector, DSIP is involved in projects ranging from machine learning applied to spatiotemporal data for weather nowcasting and forecasting, Earth System Modeling, statistical downscaling, data quality, and anomaly detection of observations. As for applications in physics, DSIP is developing Deep Learning solutions for data analysis in high-energy physics experiments and in space physics.
Integrative AI that combines deep learning and formal methods; predictive maintenance and condition monitoring; physics-informed models, differentiable surrogates and foundation models for multivariate time series.
Precipitation nowcasting from weather radar, early-warning systems, deep-learning downscaling of forecasts and climate scenarios, and better parametrisations for Earth system models.
Automatic pruning of neural networks, space weather and ionospheric monitoring, and forecasting of geomagnetic storms.
Finds where synthetic data can enable or improve AI along a space mission, with priority to ground segment engineering, operations and space safety. DSIP develops and validates methods for selected data types in minimum viable products and drafts the roadmap towards operational products.
Two AI algorithms for road safety. The first estimates whether a driver's behaviour, from floating car data, is consistent with the safe speed, using accident statistics, weather, traffic and road geometry from OpenStreetMap; the second warns the driver in real time about the speed to keep before and through a bend.
The Italian AI Factory: high-performance computing and AI services for start-ups, SMEs, public administrations and research. OpenIoT develops AI models for climate-resilient and resource-efficient agriculture.
Software components that combine machine learning with symbolic models of systems and faults to make FDIR more accurate and efficient. The goal is more autonomous and reliable aerospace systems, such as satellites and exploration vehicles, over a 24-month project.
A deep-learning model on weather-radar data forecasts thunderstorms, heavy rain, snow and hail at 500 m and 5-minute steps up to 100 minutes ahead, in under 10 seconds. The system gathers radar, satellite, lightning and ground-station data and issues automatic alerts by SMS, email, phone and the IO app.
A yearly school for students from industry and academia on AI for weather and climate.
A high-productivity environment for deep learning time-series forecasting.
A weather language model for precipitation nowcasting.
A modular real-time pipeline for machinery diagnosis and prognosis.
