PORTALE DELLA DIDATTICA

Ricerca CERCA
  KEYWORD

ELECTRONIC DESIGN AUTOMATION - EDA

Reinforcement Learning from Sim to Real: Smart Battery Warehouse Management System

keywords BATTERIES, BATTERY, DEEP LEARNING, ENERGY MANAGEMENT, MACHINE LEARNING, OPTIMISATION, REINFORCEMENT LEARNING, WAREHOUSE

Reference persons DANIELE JAHIER PAGLIARI

External reference persons M. O. M. E. Elshaigi (Comau S.p.a.)

Research Groups DAUIN - GR-06 - ELECTRONIC DESIGN AUTOMATION - EDA, ELECTRONIC DESIGN AUTOMATION - EDA, GR-06 - ELECTRONIC DESIGN AUTOMATION - EDA

Thesis type EXPEIRMENTAL, IN COMPANY, EXPERIMENTAL, MACHINE LEARNING, SOFTWARE DEVELOPMENT

Description Cell formation is the last process in the EV (Electrical Vehicles) Li-Ion Battery cells production chain, where the battery cells are activated through a series of charge and discharge with resting phases and electrical tests in between. It is considered the second top in terms of cost after the materials due to the expensive power electronics and energy consumption. A Giga scale factory operates in a magnitude of GWh/year capacity and consumes a vast amount of installed energy.

The Giga factory usually contains several warehouses (pre-formation- formation, grading ...) with different layouts and utilization purposes. Each warehouse contains many clusters (towers) with groups of dual usage (charge/discharge) formation chambers. During the formation process, a significant amount of power conversion losses (~25-30%) occur. With an optimized architecture of the formation warehouse developed by Comau, it is possible to transfer the power from a chamber in the discharging phase to another chamber in the charging phase in the same cluster.

The goal of the thesis is to build a Reinforcement learning based management system to predict the next formation chamber/s to be occupied by the next group of cells arriving at the warehouse to maximize the power recycling and reduce the overall power consumption:
- Design a data acquisition pipeline (in python) to run remotely simulation models on a cloud-based plant simulation and collect raw data.
- Construct and pre-process several datasets from the collected raw data.
- Lastly, train and evaluate an RL model, the results can be benchmarked to the optimizers used in plant simulation.

Required skills Python Programming; Machine and Deep Learning Models; Algorithms;

Notes Thesis in collaboration with the Comau E-mobility Global Competence Center (in Turin).


Deadline 15/03/2025      PROPONI LA TUA CANDIDATURA