Recursive Prediction Error Gradient-Based Algorithms and Framework to Identify PMSM Parameters Online

Abstract

Real-time acquisition of accurate machine parameters is of significance to achieving high performance in electric drives, particularly targeted for mission-critical applications. Unlike the saturation effects, the temperature variations are difficult to predict, thus it is essential to track temperature-dependent parameters online. In this paper, a unified framework is developed for online parameter identification of rotating electric machines, premised on the Recursive Prediction Error Method (RPEM). Secondly, the prediction gradient ($\mathbf{\Psi}^T$)-based RPEM is adopted for identification of the temperature-sensitive parameters, i.e., the permanent magnet flux linkage ($\Psi_m$) and stator-winding resistance ($R_s$) of the Interior Permanent Magnet Synchronous Machine (IPMSM). Three algorithms, namely, Stochastic Gradient (SGA), Gauss-Newton (GNA), and physically interpretative method (PhyInt) are investigated for the estimation gains computation. A speed-dependent gain-scheduling scheme is used to decouple the inter-dependency of $\Psi_m$ and $R_s$. With the aid of offline simulation methods, the main elements of RPEM such as $\mathbf{\Psi}^T$ are analyzed. The concept validation and the choice of the optimal algorithm is made with the use of System-on-Chip (SoC) based Embedded Real-Time Simulator (ERTS). Subsequently, the selected algorithms are validated with the aid of a 3-kW, IPMSM drive where the control and estimation routines are implemented in the SoC-based industrial embedded control system. The experimental results reveal that $\mathbf{\Psi}^T$-based RPEM, in general, can be a versatile technique in temperature-sensitive parameter adaptation both online and offline.