장단기 메모리 네트워크와 최소 평균 제곱 적응 필터를 활용한 능동 진동 제어 기법의 시스템 식별 성능 비교 연구

This paper presents a comparative study of active vibration control methods through system identification. The focus is on optimizing the transfer function coefficients of the Least Mean Squares (LMS) adaptive filter and the data-based identification and prediction algorithm for hidden state with Long-Short Term Memory (LSTM). To compare the performance of the two system identification methods, a simple low pass system is used, and both methods show appropriate system identification and prediction performance with 0.005, 0.008 RMSE value with training data and 21dB, 27dB PSD reduction compared to original test data, respectively. This paper also shows system identification and data prediction performed for the rotor hub and control surfaces of a medium-sized rotary-wing aircraft, Using ABAQUS data. The results showed that the LMS adaptive filter achieved a 0.035 RMSE value with training data and a 8.78dB PSD decrease compared to the original test data, indicating appropriate system identification and prediction performance. However, the long-term memory structure experienced overfitting during training, indicating the need for network structure optimization.