| KEYWORD |
Design of a digital twin of an electrical drive using Machine Learning
keywords AUTOMATIC CONTROL, DIGITAL TWIN, ELECTRICAL MACHINES, MACHINE LEARNING, ARTIFICIAL NEURAL NETWORKS, MACHINE LEARNING, DEEP LEARNING, OPTIMIZATION,, MATLAB, PYTHON, WIND TURBINE
Reference persons EROS GIAN ALESSANDRO PASERO, VINCENZO RANDAZZO
Research Groups Neuronics (Artificial Neural Networks)
Thesis type EXPERIMENTAL, EXPERIMENTAL AND MODELING, EXPERIMENTAL AND SIMULATIONS, EXPERIMENTAL APPLIED
Description The thesis aims at developing an innovative control methodology of electrical drives and power converters based on the Digital Twin (DT) and Machine learning and/or Artificial intelligence (ML/AI) able to provide several advanced control functions, such as health monitoring, estimation of uncertainty parameters, predictive maintenance, fault detection and management.
The real-time digital model receives data from the physical system, like operating environments, functionalities, working conditions, sensor data, and so on, through communication interfaces or protocols.
The real-time digital model processes these data, using Machine learning and/or Artificial intelligence (ML/AI), to update itself in real-time and send some control commands to provide optimization and decision support for physical systems. By continuously monitoring the components in real-time, it is possible to act with a different strategy (e.g. preventive maintenance) in the event of sudden stress for the component that would lead to sudden breakage. In addition, in the perspective of pervasive use of wide bandgap power devices (Sic and Gan), the proposed approach allows operating at much higher voltages, frequencies, and temperatures than conventional. Indeed, machine learning is a very powerful tool to recognize patterns in data
(e.g. anomalies) and to detect (diagnosis) and prevent (prognosis) machine faults. In this sense, periodic component checks can be avoided or reduced in frequency, which is a great advantage especially in offshore wind farms where the distance is one of the major costs.
The proposed methodology will be implemented and applied to a real case such as the wind offshore applications, providing the great advantage of reducing the periodic maintenance of the power conversion system. Thanks to this, periodic power converter component checks can be avoided or reduced in frequency, which is a great advantage especially in wind offshore applications where distances represent one of the major costs. The proposed method can be easily applied to other sectors, such as automotive, aerospace, railway, naval, marine sectors, where the gradual evolution from hydro-pneumatic to electrical disposition of power has placed stringent requirements on the reliability of power electronic components power conversion system.
Required skills passion for thesis work. basic knowledge of neural networks, deep learning and electrical drives can be useful. otherwise, we will provide all the material
Deadline 23/02/2025
PROPONI LA TUA CANDIDATURA