In order to improve the fault diagnosis precision of rolling mill main drive system

a fault feature extraction algorithm based on EMD-AR model and correlation dimension is proposed.In the proposed wave-let and

EMD are used to decompose the vibration signal of complex machine into several intrinsic mode function（IMF）

the AR models of some IMF components which contain main fault information are constructed.The correlation dimensions of auto-regressive parameters in AR models are calculated.Analysis results of the experimental data show that this method not only can reflect the state changes of dynamic system profoundly and comprehensively

but also can realize separation of the state features.It provides reliable basis for judging the fault conditions of rolling mill main drive system effectively.