This thesis aims to develop an innovative soft sensor for the mixing-grinding process, crucial for nuclear fuel fabrication. The ambition is to create a soft sensor powered by experimental devices, providing real-time estimation of the properties of the product.
The central goal is to design a method that accurately and at high frequency estimates the properties of the granular medium, such as grain size. This project envisions establishing a direct relationship between physical sensors and material properties through simulation and a learning model. The challenge lies in predicting these properties without directly assessing the internal state of the process.
The candidate will work at the Fuel Research Department (IRESNE Institute, CEA-Cadarache). The candidate will first inventory suitable physical sensors, before constructing a synthetic database from simulations. This data will feed a deep recurrent learning model, adjustable by a Kalman filter. The challenge also lies in effectively integrating real test bench data with simulations, while managing inaccuracies and discrepancies.