Encoder Decoder architecture with Bi-LSTM for model free sensorless control of induction motor

• Proposed a novel Bidirectional Long Short-Term Memory (Bi-LSTM) encoder-decoder for rotor speed estimation • Used the estimator in sensorless field-oriented control of induction motor • Validated estimator via simulations under various load and speed profiles • Compared closed-loop performance using estimated versus measured speed • Estimated speed feedback showed strong results in key performance metrics This paper proposes a novel encoder-decoder architecture with Bidirectional Long Short-Term Memory (Bi-LSTM) recurrent skip connections for accurate rotor speed estimation of an induction machine. The proposed sensorless controller uses stator current, stator voltage, torque, and speed reference signals, to perform time-series prediction and accurately estimate the rotor speed. The inclusion of a Bi-LSTM layer based recurrent skip connections enhance the model’s ability to learn long-term dependencies, thus significantly improving estimation accuracy under various loading conditions. The estimator is trained for both transient and steady-state behavior using a dataset that includes rotor speed trajectories in both positive and negative directions under various loading conditions. The proposed speed estimator is implemented within a field-oriented sensorless control framework for an induction motor drive system. Simulations are conducted using MATLAB/Simulink to validate the proposed estimator’s performance under various loading conditions and different speed profiles, including a rated-speed step and square-wave references. The simulation results demonstrate that the proposed estimator accurately estimates the rotor speed. Moreover, when the estimator is implemented in a field-oriented sensorless control framework, it performs comparably to the controller utilizing measured speed, particularly in key metrics such as response time, steady-state error, and integral mean square error.