PMSM Parameter Estimation Using Singular Value Decomposition 2018-01-0455

Optimal field weakening control of Permanent Magnet Synchronous Motors (PMSM) relies on accurate parameter knowledge. From a modeling perspective, a good parameter set allows better correlation between the model and physical system. Errors in inputs (parameters) such as back-emf (K_e), phase resistance (R), battery cable resistance (R_c), phase inductance (L) and voltage multiplier (K_v) have a coupled effect on the system outputs (measurements) such as torque, battery current and DC link voltage; the relationship between inputs and outputs is not one to one. A single parameter change affects more than one output or vice-versa. For example, in a torque-speed curve before the knee point (base speed), only the K_e parameter affects the output torque but after the knee, all parameters influence the torque. Also, any change in parameters can have unwanted affects like torque ripple, which can lead to NVH issues. Singular Value Decomposition (SVD), coupled with a Recursive Least Squares (RLS) based analysis technique, provides insight into the relationship between system inputs and outputs. It helps sort and weigh the parameters from most important to least important. SVD shows that by adding more measurements, the estimation process is improved and the measurement space more fully spans the parameter space. The results show that the estimated parameters are indeed accurate and provide good correlation between simulation and dynamometer results.