This paper presents a framework for fault diagnosis of complex systems by proposing a new multi-sensor and multi-domain knowledge information fusion (MSMDK) model. In the model, the multi-domain case, rule and model knowledge are adopted for multi-sensor information reasoning, and then the advanced Dempster-Shafter (D-S) evidence theory is proposed for overall fusion, which has the advantage of avoiding the evidence conflict by defining the evidence preference weight (EPW). In addition, the basic belief assignment (BBA) methods of the three different domain knowledge are brought forward for entire fusion process. The experiments of fault diagnosis of helicopter autopilot are carried out and demonstrate that the proposed method can serve as an effective method for the fault diagnosis of complex dynamic systems.