A dynamical system analysis method is presented, that permits the characterization of unsteady phenomena in a centrifugal compression system. The method maps one experimental time series of data into a state space in which behaviors of the compression system should be represented, and reconstructs an attractor that geometrically characterizes a state of the compression system.
The time series of data were obtained by using a high response pressure transducer and an analog to digital converter at surge condition. For the reconstruction of attractors, a noise free differentiation method in time was employed. The differentiation was made by high order finite difference methods. To remove the influence of noise, the data were passed through a filter using a third order spline interpolation. In this study, the dimension of the state space was restricted to three. The measured data itself and their first and second derivatives in time are used to represent an attractor in the state space.
The modeling of the system behavior from the time series of data by second order ordinary differential equations was attempted. It is assumed that the data and their derivatives satisfy the equations at each time. Then, appropriate coefficients are determined by a least square method.
The reconstructed attractor showed complex cyclic trajectories at a first glance. However, by applying a band pass filter to the original signal, it was found that the attractor consisted of three independent wave forms and formed an attractor with torus-like behavior. In contrast, the solution by the modeled equations showed a type of limit cycle.