Method for Thermal Property Characterization From In Situ Measurements Applied to the Performance Evaluation of Electric Machine Windings

A combined experimental and model-based methodology is proposed for characterizing thermal properties of electric machine components, integrating assembly-level measurements, inverse modeling, and uncertainty quantification. Unlike traditional approaches, the method applies sensor weighting and assesses parameter identifiability, supporting improved measurement design. The methodology is demonstrated by quantifying maximum current density of five wire types—rectangular copper and aluminum, thick and thin stranded, and litz wires—using motorette setups. The proposed methodology minimizes the weighted mean absolute error between model and experiment across all motorettes from 12.9 °C (literature properties) to 1.2 °C. Global uncertainty analysis reveals limited identifiability of core axial conductivity and winding–wedge interface resistance. Current densities, including skin and proximity effects, are simulated using fitted thermal properties. At low frequency, copper rectangular wires perform best (8.9 A/mm²), followed by aluminum (6.9 A/mm²), stranded thick (6.3 A/mm²), and stranded thin and litz wires (6.2 A/mm²). At 1 kHz, stranded thin and litz wires perform best (6.2 A/mm²), while aluminum rectangular (4.8 A/mm²), copper rectangular (4.2 A/mm²), and stranded thick wires (3.9 A/mm²) show reduced performance. Using literature thermal properties yields 1–13% lower current densities. These results validate the methodology and highlight its value for uncertainty-aware thermal characterization, sensor placement, and quantitative electromagnetic–thermal wire performance assessment in machine design.