In order to solve the problem that it is difficult to determine the basic probability distribution value in the evidence fusion of petrochemical rotary unit fault signals, this paper proposes a dimensionless index-based hybrid approach using the two-sample Kolmogorov-Smirnov (K-S) test. This method decreases the effect of the interference from the fault characteristic sensitivity and operating conditions through the dimensionless index's calculation. It can also find out the evidence needed for fusion basic probability assignment values by the two sample K-S test. Finally, it conducts evidence fusion using a modified D-S algorithm. The advantage of the proposed method that makes it superior to other methods is that it can deal with fuzzy, incomplete or accurate data. Therefore, the input signals of petrochemical rotary machinery equipment and sensor equipment can be processed directly. In this paper, experimental results prove the effectiveness of the proposed method, so the decision risk can be reduced.