Estimating fill factor and equivalent insulation thermal conductivity in AC contactor coils using hotspot temperatures and Bayesian neural networks

Abstract

This study introduces a Bayesian neural network (BNN)–based framework for non-invasively estimating two fundamental insulation-related parameters of AC contactor coils—the fill factor and the equivalent thermal conductivity—using measured hotspot temperatures as the primary diagnostic input. Together, these parameters play a crucial role in determining the coil’s temperature rise, thermal stability, and long-term reliability. However, accurately determining them through direct measurement is often difficult or impractical, particularly for compact contactor coils where disassembly can damage the winding or is impossible without specialized instruments. To address this challenge, the finite element method (FEM) is employed to model the coil’s temperature distribution under varying insulation conditions, thereby generating a comprehensive and physically grounded dataset for training the BNN. In the proposed method, hotspot temperatures—obtained indirectly from measured ambient conditions and coil surface temperatures—serve as the inputs from which the BNN predicts the corresponding fill factor and equivalent thermal conductivity. By eliminating the need for coil teardown or complex laboratory procedures, this approach offers a practical, data-driven solution for evaluating coil design, diagnosing insulation characteristics, optimizing thermal management strategies, and ultimately enhancing the reliability and performance of electromechanical switching devices.