In recent years, reusable launch vehicles have been pursued globally to realize low-cost and high-frequent space transportation. For such reusable launch vehicles, "Health Monitoring System" that constantly monitors the engine operating performance and detects and identifies the location and cause of a failure in real-time during flight is required to return safely to the ground. In this report, a development method of a failure diagnostic system using simulation data for a reusable rocket engine is considered and demonstrated. The failure diagnostic system contains anomaly detection by Mahalanobis distance and failure classification by support vector machines to enable identification of both known failure modes and unknown failures. The machine learning of the failure diagnostic system was performed with simulated operation data only produced by "Plant simulator" of the rocket engine, due to the difficulty of collecting sufficient actual operating data of engine firing. The prototype failure diagnostic system was developed and demonstrated the capability with actual rocket engine firing test data. It was confirmed that the failure diagnostic system trained only by the simulation data can detect and classify a failure which occurred in the actual engine operation.