Abstract: This paper deals with a method to identify troube of a machine by its vibration noise. At first vibration noise of the machine is tramformed from time domain into fre-queney domain using Fourier trasfom. Its logarithmic power spectrum is N_xx(f). Then define spectrum correlation coefficient:r_f(0)\text=\frac\int _f_2^f_2N_xx(f)N_yy(f)df \left\int _f_2^f_2N^2_xx(f)df\int _f_1^f_2N_yy^2(f)df \right^1/2
Befor correlation operation, normalize N_xx(f) and N_yy(f) with their average, and subtract "DC component" 1.
It follows that the spectrum correlation coefficient of normal machines themselves is larger than the spectrum correlation coefficient between the normal and the trouble. The spectrum correlation coefficient of different troubles of a machine is smaller than same trouble, but larger than between the normal and the trouble.
If a certain thleshold is selected (threshold of spectrum correlation) the trouble will have been discriminated.