Study On Rocket Engine Turbo Pumps Real-time Fault Detection Method

With the development of China’s manned space project, the ever-increasingdemands are put forward for the launch vehicle propulsion systems.Studying on the newlarge-thrust carrier rocket is an important part of the aerospace engineering,whilecarrying out the fault diagnostic research of the carrier rocket propulsion systems, is themost important one.As one of the core part of the liquid rocket engine, the turbopumphas been the focus of fault detection.To carry out the study of the turbopump real-timefault detection method is much valuable for the development of a new launch vehiclepropulsion system.Currently, real-time fault detection algorithm based on the vibrationsignal is a major research direction of the real-time status monitoring of high-speedrotating machinery.This paper studies the turbopump vibration signals of real-time faultdetection,analysis and comparison of the commonly used statistical features for thevibration fault signal features linear correlation between stability and sensitivity.Todetermine the follow-up study on the basis of real-time fault detection should select therms, the margin factor and kurtosis as the feature amount of the fault detection. Thenproposed turbopump real-time fault detection accuracy and real-time evaluation criteria,and the historic commissioning of a certain type of liquid rocket engine turbopumpvibration data of three categories based on the time domain, frequency domain andsupport vector machine algorithm.Real-time fault detection algorithm based onHilbert-Huang Transform. Algorithm Hilbert Huang Transform, the vibration signalwith EMD empirical mode state function IMF to consider the requirements of real-timefault detection and selected characteristics HHT transform spectrum peak as the amountof fault feature detection。On the basis of the research SVM principle, put forward areal-time fault detection algorithm based on support vector machine (SVM), the use ofthe fault sample points rms as the characteristics of the fault detection and optimizationto improve the training speed and detection efficiency, to achieve detection of the failureof the historical test data。To study classical algorithm based on time-domain analysis.Red line alarm algorithm by studying multi-feature, multi-feature adaptive algorithm and adaptive threshold algorithm three classic fault detection algorithm, and examinetheir effect on the real-time fault detection and as other reference algorithm.On this basis, the use of test data of a certain type of liquid rocket engineturbopump history, experimental verification of the algorithm described above to obtainthe relevant data and the detection result of the fault detection. The last test results andthe accuracy of the above algorithm, real-time comparative analysis. Several algorithmsare able to accurately detect a fault, but in real time on a remarkable performance merits.Which adaptive threshold algorithm based on Hilbert Huang Transform algorithm basedon support vector machine fault detection algorithm can meet the accuracy and real-timerequirements.