Parametric and non-parametric models for lifespan modeling of insulation systems in electrical machines

International audienceThis paper describes an original statistical approach for the lifespan modeling of electric machine insulation materials. The presented models aim to study the effect of three main stress factors (voltage, frequency and temperature) and their interactions on the insulation lifespan. The proposed methodology is applied to two different insulation materials tested in partial discharge regime. Accelerated ageing tests are organized according to experimental optimization methods in order to minimize the experimental cost while ensuring the best model accuracy. In addition to classical parametric models, the life-stress relationship is expressed through original non-parametric and hybrid models that have never been investigated in insulation aging studies before. These two models present the original contribution of this paper. For each material, models are computed from organized sets of experiments and applied on a randomly configured test set for validity checking. The different models are evaluated and compared in order to define their optimal use