Fault Detection And Diagnosis Of Liquid Rocket Engine Based On Data Mining

Liquid rocket engine is widely used in aerospace launch. It has been the most possible fault-occurring system under difficult working conditions. By statistics, more than 50% of critical failures were caused by fault of propulsion system. So the elevation of the reliability for propulsion system is the key point. In the test of rocket engine, there are huge amount of data to be analyzed. To find out new knowledge from the test data, furthermore, to discover the deep regularity of rocket engine, is an important project.Data mining is an integrated application for statistics, data-base, pattern identification, artificial intelligence, and visualization technique and data analysis. It is a process to find valuable information and knowledge from a lot of incomplete, noisy, vague, stochastic date which is hidden or unknown to people. This thesis tries to apply data mining methods to liquid rocket engine to effectively detect fault and find out the knowledge underneath. Take some effective measures to reduce the time of test running, to shorten research period and to save the research expense, so as to minimize the loss caused by engine faults.Quantitative association is developed by the basic of Boolean association. It transforms numeric and multiple properties into Boolean properties. Based on the theory of quantitative association rules, numeric data of a certain large-scale liquid propellant rocket engine were divided into intervals with statistical method, and tested with FP-Growth algorithm, which verified the feasibility of the method.Classification is one of the important parts in data mining. Bayesian classifier is a typical classifier based on statistics. It can forecast the possibility of class membership, for example, calculating the probability of a sample belonging to a special class. Bayesian classifier uses training set to conclude the classifier, and uses the classifier to classify the rest data. In this thesis, two typical Bayesian classifiers, Naive Bayesian classifier and TAN (Tree-Augmented Naive Bayesian Network) classifier, were used to classify the faults of liquid propellant rocket engine, and hot test data and simulated data were used to contrast and verify the classification precision.Liquid rocket engine fault detect and diagnosis by quantitative association and Bayesian classifier were completed in the thesis, and the two parts were integrated into the data mining system—"Liquid Rocket Engine Fault Detect and Diagnosis Data Mining System". On the base of previous works, a simple voting system which can display the result of each system was added. The system becomes more significant to the fault diagnosis of liquid rocket engine.