Characterization of nonlinear dynamic systems using artificial neural networks
- Univ. of Texas, El Paso, TX (United States)
- Los Alamos National Lab., NM (United States). Engineering Science and Analysis Div.
- Sandia National Labs., Albuquerque, NM (United States). Experimental Structural Dynamics Dept.
The efficient characterization of nonlinear systems is an important goal of vibration and model testing. The authors build a nonlinear system model based on the acceleration time series response of a single input, multiple output system. A series of local linear models are used as a template to train artificial neutral networks (ANNs). The trained ANNs map measured time series responses into states of a nonlinear system. Another NN propagates response states in time, and a third ANN inverts the original map, transforming states into acceleration predictions in the measurement domain. The technique is illustrated using a nonlinear oscillator, in which quadratic and cubic stiffness terms play a major part in the system`s response. Reasonable maps are obtained for the states, and accurate, long-term response predictions are made for data outside the training data set.
- Research Organization:
- Sandia National Labs., Albuquerque, NM (United States)
- Sponsoring Organization:
- USDOE Office of Financial Management and Controller, Washington, DC (United States); USDOE, Washington, DC (United States)
- DOE Contract Number:
- AC04-94AL85000; W-7405-ENG-36
- OSTI ID:
- 291159
- Report Number(s):
- SAND--98-2135C; LA-UR--98-2945; CONF-981031--; ON: DE99001064; BR: YN0100000
- Country of Publication:
- United States
- Language:
- English
Similar Records
Application of an artificial neural network to reactor core analysis