Data-driven approach to observer-based incipient fault detection in transformers

Detection of incipient faults in transformers represents a very important issue because of the maintenance and in order to avoid further failures. There are many approaches to detect incipient faults off-line, and only a few considering the possibility to realise the detection on-line. This work considers an on-line incipient-fault detection from electrical measurements of electrical transformers. The currents are used together with a data-driven algorithm to obtain the design of a diagnostic observer. The observer is further used to supervise the behaviour of a transformer. Simulation results show viability the proposed approach.