Abstract: For some processes in which the mean values of process variables vary in different phases, a new fault diagnosis approach based on improved principal component analysis (PCA) was proposed. The state of process variables are transformed by a high-pass filter for system extension, then PCA is applied to the output of the extended system to develop a statistical model, by which the process monitoring and fault diagnosis are both available. This method can eliminate the negative effect of mean value change on the conventional PCA model and improve further the robustness and sensitivity of fault diagnosis. The method was applied to diagnosing the fault of cooling water leakage of the system of vacuum consumable electric-arc (VCEA) furnace, and the simulation results showed that the proposed method is effective.