Gauss-Newton: A prediction-error-gradient based algorithm to track PMSM parameters online

Abstract

Online adaptation of temperature-sensitive motor parameters is of significance for the electric drives in reliability-critical applications. Recursive prediction error method (RPEM) is widely used for this purpose. Gauss-Newton Algorithm (GNA), a prediction-error-gradients based algorithm, is adopted in this paper to find RPEM-gains for the parameter identification. This paper first investigates the simultaneous identifiability of permanent magnet flux linkage ( Ψm), and stator resistance (Rs) of interior permanent magnet synchronous machine (IPMSM) using both nonlinear observability theorem and RPEM. Subsequently, GNA is analyzed for its tracking capability, speed of convergence, need of gain-scheduling and computational demand in comparison to stochastic gradient (SGA), another algorithm of the same class, using steady and dynamic state simulations.