DC-2: Niklas Groll

Contact

E-mail: nigro@kt.dtu.dk

Project

Hybrid approaches for smart integration of AI/deep learning with model-based algorithms for online risk monitoring

Host organization

DTU

Supervisors

Prof. Gürkan Sin (Main, DTU); Prof. Alessandra Russo (co-supervisor, IMPERIAL)

Duration

36 months

Objectives

The overall objective is to develop a new hybrid modeling framework for risk monitoring that combines model-based algorithms with AI (especially deep learning) algorithms through the integration of mechanistic knowledge encoded in first principles models (process simulation technologies) with the empirical yet efficient data-driven approach of big data analytics.

A proof concept integration of tools will be achieved by interfacing big data analytics tools in high-level programming environments with model-based tools in a process simulation environment for effective and synergistic implementation. The framework will be applied and validated on selected case studies from the urea industry during the secondment with an industrial partner (NOVOTEC).

DC-2 will work in close collaboration with DC-3 (for model-based) and DC-4 (for data-based) algorithms.

Expected results

  1. Development of framework based on hybrid models integrating model-based methods with advanced machine learning algorithms,
  2. Development of a dynamic model for prediction of flaring and training with reconciled data,
  3. Application of Kalman filter for reconciled data in close collaboration with DC-3 and an industrial partner (NOVOTEC) for urea industrial case study.

Planned secondments

  1. IMPERIAL, M12, 2 months: Training on deep learning and artificial intelligence algorithms for online risk monitoring 
  2. NOVOTEC, M30, 3 months: Deep learning case study training and application for urea operations