本文主旨為研究時間延遲(Time-Delay)效應對結構主動控制系統的影響,並建立一類神經網路的預測機(Predictor),預測結構反應以修正控制系統,進行時間延遲的補償(Time-Delay Compensation),確保結構主動控制系統受到時間延遲影響下的穩定性。首先建立模糊神經網路的主動控制系統,並在此控制系統前,設置一神經網路預測機,修正控制系統的輸入,進而獲得時間延遲補償所需的控制力。 時間延遲的補償方法,一般常採用的方法有下列兩種:一是利用時間延遲產生振動相位移位(Phase Shift)的觀念,修正系統回饋曾益矩陣,以達補償效果。但此法需事先瞭解結構的振頻與延遲時間。另一方法為線上求解運動方程,雖此法必須進行大量的計算,但預期效果較佳。本文即為此架構下建立神經網路預測機,進行時間域的補償,並與振動相位角移位(鍾立來,八十三年)的補償方法進行比較。模擬的結果:單自由度在延遲5步(step,取樣時間為0.02sec)內兩法控制效果接近,但延遲神經網路補償的效果較振動相位角移位法不管在位移反應或控制力上都有較佳的效果。在三自由度模擬上,受到模態影響,因此時間延遲1 步(step)下可發揮補償效果。 關鍵字:時間延遲、相位移位、時間延遲類神經網路
The purpose of this research is to study about time delay effect in active structural control. If time delay is occur in control system, then we cannot ensure not only the control system is stable but also the structure response convergence. We attempt to establish a neural network predictor in order to predict the state of structure, and modify the controller to compensate the time delay effect. There are two time delay compensations being used usually. One is phase shift compensation that should be identified the mode of structure and delay time. Another is solving the equation of motion on-line. This method is more efficient despite of huge number of calculations. Two kinds of controller were selected as follows: one is variable structure control system (VSS). Another is fuzzy-neural network (FNN). And two time delay compensation ways of phase shift and time delay neural network. Thus, we have two kinds of controller and two kinds of compensation ways and four combinations. Each of them are selected and compared with others. The results of numerical simulation in single degree of freedom (SDOF) structure: if the controller is FNN with delay time less than three steps, both of phase shift and time delay neural network predictor could compensate the time delay effect efficiently. On the other hand, if the controller changed to VSS with delay less than fifteen steps, both of two compensation ways could make the response become lower. However, the time delay neural network predictor is better than phase shift compensation in response of displacement, velocity and control force. In the case of three degree of freedom structure: system delay one step may cause the response of structure diverge. Phase shift compensation and time delay neural network predictor can decrease the displacement and velocity response in system delay one step. Keywords: time delay, phase shift, TDNN, neural network