Studies On The Application Of Data Mining For Fault Detection And Diagnosis Of Liquid Rocket Engine

The faults of liquid rocket engine are all kinds of abnormal states and the states different from preconcerted working states. The fault detection of liquid rocket engine is a judgement process for engine working state by measured data. The fault diagnosis of liquid rocket engine can be considered as a problem of mode identification, the task of which is to divide samples into mode types by obtained characteristics. Liquid rocket engine is a thermal kinetic system, it runs under extremely strict physical conditions, such as high temperature, high pressure, strong erosion, high density of energy release, strong components parameters coupling. It is very difficult to establish accurate dynamic mathematical model. Furthermore, the established model is often high-level and non-linear owing to the complicated working process of the engine. The application of the fault detection and diagnosis algorithms based on mathematical model and simulation for general dynamic system is very difficult.A lot of parameters needed to be detected for the fault detection and diagnosis of liquid rocket engine. Data need to be recorded at every certain time, so, large amount of data need to be analyzed. However, the valuable information hidden behind of the data is not mined and utilized effectively. Data mining techniques are the merged result of many disciplines drived by application demands. It is a comprehensive application of statistics, data warehouse techniques, mode identification, artificial intelligence and visualization techniques. It is a process to find valuable information and knowledge hidden behind data and unknown from a lot of incomplete, noise, vague, random, real application data. Data mining promotes data owner's comprehension, cognition and application for the large amount of origin data by finding new rules and new conception. In this paper, data mining was applied to acquire knowledge to solve the bottleneck problem of knowledge acquisition in the fault detection and diagnosis of liquid rocket engine.Rough set theory is a powerful tool in dealing with vague and uncertain information. In this paper, by analyzing the fault characteristics of liquid rocket engine and the rough set method, rough set theory was applied to mine the hot-firing test data of liquid rocket engine. By investigating 4 groups of test data of a certain type of engine, 24 original attributions were reduced to 2 groups, each group had 2 attributions, Critical parameter combinations for fault diagnosis for liquid rocket engine were obtained; 213 records were reduced to 5 or 6 records.Association rules were used to discover association relations between sets of items in large database. The fault of liquid rocket engine in the starting-up and shutdown process could be detected by judging whether the rules were valid or not. By applying the method to the hot-firing test data of liquid rocket engine in the starting-up and...