abstract
- Accurate angular speed estimation is critical for controlling mechanical and electromechanical systems. Linear Hall sensors are widely used for this purpose, but their signals are often distorted by magnet misalignment and manufacturing imperfections. These distortions introduce nonlinear effects that traditional filters and observers cannot handle effectively, leading to poor estimation performance. Conventional models often rely on compensation mechanisms to correct these distortions, but such strategies are limited and reflect the inability to fully capture the system's true behavior. This paper proposes a novel AI-powered strong observer using a single-layer Long Short-Term Memory (LSTM) neural network. The method maps magnetic field measurements directly to angular speed, avoiding the need for compensation, explicit modeling, or prior knowledge of distortion characteristics. Trained on real-world data, the observer inherently captures nonlinearities and signal dynamics, providing robust and accurate estimation across a wide range of operating conditions. Experimental results confirm the effectiveness and practical advantages of the proposed approach in AI-based control and estimation systems.