Based on the concise and informative causality structure obtained by a multi-signal model and the source fault information extracted by a blind source separation algorithm, a novel and effective method for diagnosing compound faults is developed. For the redundant testing signals and the multiple fault ambiguity groups when applying the multi-signal model for compound fault diangosis, a blind source separation algorithm is integrated into the multi-signal model to obtain source fault information, and the causality structure of the multi-signal model is then reconstructed. It’s analyzed theoretically that the proposed method has good diagnosticability for compound faults. The results of applying the proposed compound fault diagnosis method to a hydraulic automatic gauge control(AGC) system in a rolling process show that the diagnosis accuracy rate for all simulated double or multiple compound faults can be 100%.