A fault diagnosis monitoring method based on evidence theory and imprecise probabilities is proposed for vibration divergence fault in systems. Signals acquired by multi-sensors are used to obtain characteristic parameters. Basic belief assignments of fault are obtained by using exponential function as evidence generating function

and they are divided into two groups

which are high reliability group and low reliability group

and fused by Dempster combination rule respectively

the fused results make up probabilily interval

and calculate expectation interval which diagnose cost functions. Two different decision criterion under imprecise probability theory are used to decide whether fault is happened. The fusion diagnosis monitoring process is realized on platform of vibration isolation. The experimental results show that the proposed method can fuse information of sensor effectively and find vibration divergence fault in time.