The paper addresses the problem of predictive maintenance in high power supply systems. According to FMECA (Failure Mode, Effects and Criticality Analysis) study key issues for power system reliability are the faults occurring in 3-phase high power transformers. With respect to the state of art where typically only thermal and electrical faults are monitored this work proposes a distributed network of measuring nodes where also vibration-based mechanical stress diagnostic is implemented.
Mechanical degradation is tracked through multichannel vibration measures with sensitivity down to 0.5 mg and local processing in the transformed frequency (100 Hz-1 kHz) domain. A compact hardware-software implementing platform is also presented whose performances are assessed through experimental measurements on real high power transformers used in railway applications.