Doğrusal aktüatör olarak SMA'nın doğrusal olmayan kontrolü

Precision force control is critical for various actuator applications. Shape Memory Alloys (SMAs) are attractive candidates due to their high power-to-weight ratio and ability to generate substantial force. However, inherent complexities in SMA behavior, including non-linear relationships, hysteresis, and temperature dependence, hinder precise force control. Most existing studies primarily focus on characterizing SMA behavior and offering control methods in laboratory environments. The properties of the SMA used in real-world applications may differ from those observed in the laboratory due to changing test and operating conditions and material properties. This gap necessitates a more practical approach for reliable SMA actuator performance. This study proposes a novel approach utilizing Recurrent Neural Networks (RNNs) and real-time measurements to achieve precise force control in SMAs. A compact testbed equipped with sensors acquires voltage, current, and generated force data. RNNs are well-suited for learning complex temporal relationships, making them ideal for modeling the dynamic behavior of SMAs. Trained forward and inverse RNN models predict force output or adjust input voltage for a desired force, respectively, in real-time. This approach significantly improves force control accuracy, achieving a desired force trajectory at the actuator output with a high degree of success (e.g., 99%). This paves the way for more reliable SMA actuator operation in diverse real-world scenarios.