Inverse Thermal Modeling to Determine Power Losses in Induction Motor

Abstract

An electrical motor's inverse thermal model formulates the precision with which its power losses can be predicted from noisy local temperature measurements. This paper proposes constrained linear least square optimization followed by noise smoothing as a method to accurately predict a 37 kW induction motor's power loss components. The anisotropic heat transfer and complex geometry of the motor are represented well by the motor's lumped thermal network and 3-D finite element models, which characterize its intrinsic heat-transfer processes. With the inverse solution method, satisfactory temperature to power inverse mapping was achieved for different sets of noisy temperature inputs simulated from analytical and numerical forward solutions. This paper demonstrates that even with measurement noise in input data, if reliable information about the expected response of the system is available, an accurate reconstruction of power losses can be achieved.