Cascading failures in critical cyber physical systems such as power systems are rare but lead to huge social and economic implications. Timely diagnosis of faults in these systems is a challenging task due to inherent heterogeneity and scale of the system. In the past, we have successfully demonstrated a robust technique for diagnosing independent component faults using Temporal Causal Diagrams (TCD) at sub-system level. In this paper, we present a systematic approach of using the sub-system level fault models to auto-generate a systemlevel fault model that helps in diagnosing cascading failures. We show the time complexity of our model generation algorithm using industry standard Power Transmission networks. Further, we describe the updates to the existing TCD reasoner algorithms and report the TCD diagnosis results for simulated multi fault scenario on a standard power system.