A PINN-Based Lifetime Predictor for Oil-Impregnated Paper Insulation Degradation in Power Transformers Using Degree of Polymerization

This paper investigates the aging behavior of oil-impregnated paper insulation in power transformers by estimating the degree of polymerization (DP). The paper proposes an efficient and reliable physics-based lifetime estimation model cou-pled with artificial intelligence techniques. The RUL estimation is important for the industry to develop the power transformers' condition-based maintenance plan and to avoid operation inter-ruption in the power systems. As a result, this approach will help transition the prognostics and health management system from a 'fail and fix' strategy to a 'predict and prevent' strategy. The physics-informed neural network (PINN) model is proposed to predict the DP of the insulation paper over a certain operating time considering an additive noise to the input measurements and two types of insulation papers (non-thermally upgraded paper and thermally upgraded paper). The simulation results are carried out to demonstrate that the proposed PINN offers an improvement in RUL estimation relative to the conventional neural network without the inclusion of the physical model. Furthermore, the proposed solution is predicated on using the model that achieves the optimal solution with the least possible number of learning epochs.