PGDA – Physics Guided Data Analysis

Project description

PGDA – Physics Guided Data Analysis

The fast-paced digitalization of the physical world offers novel opportunities for a better understanding of complex physical systems. This includes data generated in global and local natural systems related to a multitude of areas, e.g., climate, meteorology, ice sheet properties and natural disasters. But equally, so data generated by human activity or human-made systems, on the ground, in the air and at sea – over and under the surface.
Obtaining these physical data via a variety of sensors and platforms, and subsequently analyzing them, requires a synthesis of physics and data science: A synthesis of physics-based modelling, simulations and analytics with statistical and machine learning (ML) based data analytics. While this is paramount to analyzing physical data, such a synthesis also provides a road to developing efficient, ML-based systems turning massive amounts of raw data from multiple sources into interpretable, actionable data and tangible insights.
This emerging field referred to as e.g., physical analytics (Hamann 2015), physics guided neural networks (PGNNs) (Karpatne 2017) or physics guided ML offers common ground for innovation in business applications – across the space, defence and security domains – and in basic science where physical data are ubiquitous.

The project has 3 goals:

1) Develop the framework for physics guided data analysis, aiming for the ability to ingest data from multiple sensor sources, normalizing data using a common taxonomy and preparing data for processing using ML.
2) Develop the framework for a physics guided ML analytics system, capable of processing normalized sensor data to provide large scale ML-driven anomaly detection and provide interpretable, actionable data and tangible insights.
3) Serve as a feasibility study for a joint large scale (DKK 50-100 mio.) and national research and development project, entailing the development of a physics guided ML analytics system based on platforms and sensors deployed in the Arctic region

The purpose of the national research and development project is the development of a physics guided and ML based data analytics system, the integration of a number of remote sensors and the deployment on a dual-use Medium Altitude Long Endurance (MALE class) UAV in the Arctic area.

Project leader

Project participants