The change of single bolt preload made little difference to the vibration mode of labyrinth disc, and the assembly state of rod fasten rotor could not be detected effectively through the time domain and frequency domain analysis of the vibration response signal. Aiming at the problem, the fractal method was adopted as a new method to analyze the assembly vibration signals of rod fasten rotor. The vibration response signals were obtained by the assembly vibration detection test. Then the fractal theory was used to analyze the test result. And the rectangle box dimension and multi-fractal spectrum parameters of three bolt preload state were figured out. The multi-fractal spectrum parameters were applied as the feature vectors of support vector machine, by which the assembly state of the rod fasten rotor was identified. The result shows that when the bolt of rod fasten rotor loosing, the box dimension of the vibration response signal will increase accordingly. When the box dimension is bigger than 1.25, it shows that the rotor assembly is unqualified. When the support vector machine is used for classification, the correct ratio reached 93.7008% while the multi-fractal spectrum is taken as the feature for assembly state prediction.