THE ONLINE SPSG PARAMETERS PREDICTION

Duration

This work proposes an online parameter-prediction framework for SPSG machines that fuses high-fidelity FEM magnetic models with a d–q axis electrical model to produce robust, fast, and physically-consistent estimates of key machine parameters (flux-linkage coefficients, d/q inductances, stator resistance, and saturation characteristics). FEM is used offline to generate parameter priors (including saturation curves and temperature sensitivity) and confidence bounds; these priors are fed as constraints and regularization (endorsement) to a real-time estimator based on a dual/extended Kalman filter (DEKF/EKF) and constrained recursive optimization operating on the d–q model. The hybrid approach maintains accuracy at low excitation (where observability is poor) by relying on FEM priors and switches to data-driven adaptation when excitation is sufficient. Results on an SPSG test rig and simulations show fast convergence and reduced bias versus purely online methods