Accelerated Degradation Data-Driven Lifetime Prediction Model of Electrical Machine Groundwall Insulation Based on Microscopic Image Characteristics

For electrical machine groundwall insulation (GI), degradation state representation (DSR) and degradation trajectory modeling are two main challenges for its accelerating-degradation-based remaining useful life (RUL) prediction. For the former, The RUL prediction accuracy based on GI’s electrical parameters is not satisfactory due to their weak correlation with GI’s residual breakdown voltage (RBV). For the later, GI’s multi-phase degradation characteristics bring great difficulty for its accurate lifecycle degradation model. Aim at solving these two problems, this article proposes a novel GI’s DSR method based on the texture and color of GI materials’ microscopic images after different aging time. This method adopts nine grayscale characteristics to establish their quantitative relationship with RBV by using Random Forest algorithm. And then, a two-phase accelerated degradation model based on Wiener process for RBV is developed to describe the degradation trajectory of GI’s RBV. Finally, this article validates the prediction accuracy of the proposed model based on accelerated degradation testing (ADT). The results show that the proposed model significantly improves the prediction accuracy of the GI’s RUL of the electrical machine.