Intelligent Online Sparse Bayesian Learning for Encoderless Motor Control With Modified Observation Reaching Law

Encoderless algorithms for surface-mounted permanent magnet synchronous motors addresses the installation limitations of encoders in harsh or confined environments. However, existing encoderless control methods suffer from challenges in dynamic response speed and parameter sensitivity issues. To solve these issues, a modified reaching law for nonlinear flux observer with improved response speed is proposed in this study. First, a logarithmic function-based reaching law is designed for the nonlinear angle observer to achieve faster response speed. The Lyapunov function is developed to prove the stability of this reaching law. Second, an accurate mathematical model of the motor is designed and a dataset is constructed based on the deviation between actual values and reference values. Adjustable weight factors are used in this model to fit factors, such as inductance saturation and unmodeled disturbances. In addition, the sparse Bayesian learning algorithm is employed to analyze the features of historical data within the dataset, which helps reduce the impact of parameter variations on angle observation accuracy. The experimental results indicate that the proposed algorithm recovers from speed reversal in only 53.2 ms, significantly faster compared to 177.2 ms in conventional methods.