Abstract: Modern nonlinear theories such as wavelet theory, fractal and chaos theory are used to analyze and study malfunction of turbo unit rotor synthetically. Three time domain vibration signals of a 28 MW turbo unit at different running state are researched by using wavelet packet decomposition and reconstruction technique and by calculating their correlation dimensions and largest Lyapunov exponents, reconstructing their phase trajectories and Poincare sections attend by. The research result shows that wavelet packet decomposition and reconstruction technique is powerful in interference elimination of signal and extracting saltation signal from complex vibration system. Correlation di-mensions or largest Lyapunov exponents of the time domain vibration signal are different obviously to the generator rotor at the three unlike running states. So conclusion which correlation dimensionality and largest Lyapunov exponent can be used as synthesis quantify index to depict mechanical fault characteristic of generator set is educed. The research can be generalized to any nonlinear dynamical system and gives a new approach for monitoring condition, fault diagnosis and prediction online.

Key words: Chaos, Fault diagnosis, Fractal, Lyapunov exponent, Nonlinear dynamics, Turbo-unit, Wavelet analysis