In the industrial process closed-loop control system, due to the adjustment function of the controller, the fault characteristics of the actuator are covered and interfered to a certain extent, and the single diagnosis method always has the phenomenon of misjudgment. To solve the above problems, this paper proposes a diagnosis algorithm based on evidence fusion. Firstly, it uses a method based on signal analysis to calculate indicators that represent fault characteristics. And they are improved in response to the “one-vote veto” phenomenon. Then, the DS evidence theory is used to fuse the probabilistic classification features based on the least square support vector machines(LS-SVM), which achieves complementary advantages. The failure mechanism information expressed by indicators is combined with the data feature information mined by probability classification, which circumvents the limitations of a single method and improves the accuracy of diagnosis. Experiments based on the dual-capacity water tank system show that this method can effectively learn the fault data characteristics of the actuator in the closed-loop system to improve diagnostic ability, and overcome the misjudgment problem of a single method, which has high application value.