Nuisance attribute projection for system independent fault diagnosis of synchronous generator

Condition based maintenance (CBM) has become increasingly important in current maintenance paradigm because of its cost effectiveness and reduced downtime. Fault diagnosis is an integral part of CBM, which requires measurement of electrical/mechanical signals under different conditions to capture the intelligence about health of the system. In system dependent fault diagnosis, diagnosis model is trained using data collected from one machine, and it can be tested on the same machine. In system Independent fault diagnosis, diagnosis model is trained using data measured from one machine, and it can be tested on machine with different capacity. In factory environment with system independent fault diagnosis, group of machines having same characteristics and different power ratings can be diagnosed using a single fault diagnosis model. In this work, we model system independent diagnosis of stator winding inter-turn faults for synchronous generator with the use of two synchronous generators having power ratings of 3 kVA and 5 kVA respectively. System dependent attributes deteriorates performance of system independent fault diagnosis, as compared to system dependent diagnosis. To improve robustness of system independent fault diagnosis model, effect of system dependent attributes should be suppressed. We use nuisance attribute projection (NAP) to improve robustness of system independent diagnosis model by diminishing the effect of system dependent features. Using NAP, performance is improved by 19.63%, 8.31% and 11.80% for R, Y and B fault models respectively, as compared to baseline performance.