Reaserch Of Turbopump Faults Detection Algorithm Based On SVM

Turbopump is the core component of liquid rocket engine with high development costs. Extreme operating environment of turbopump leads to its higher failure rate. Therefore, the research of turbopump fault detection technology is significant to reduce the failure loss of turbopump. Nowadays, the fault detection algorithm based on vibration signals of turbopump housing is a research hot topic.At first, this paper summarized and expounded the domestic and foreign technology research status of fault detection technologies of liquid rocket engine, and related theory on turbopump fault detection algorithm. Based on this, taking the case vibration acceleration signals of history test of a certain type of liquid rocket engine’s turbopump as the research object, two kinds of turbopump fault detection algorithms based on SVM(Support Vector Machine) were researched respectively. The first algorithm is based on the time-domain features and fast SVM, takes energy of sample step signals and absolute value of energy change as the time-domain features. Meanwhile, in order to solve slow training or unable training caused by too many training samples, fast SVM approach is introduced, border training sample set is screened from the original training sample set, it greatly reduces the training time while ensures accuracy of the decision-making classification function. Meanwhile, a multi-index weighted alarm strategy is proposed, it provides a basis for using the algorithm to judge and detect whether the step signals contain failures. The second algorithm is based on frequency-domain features and fuzzy classification SVM. The algorithm takes frequency amplitude standard deviation of the sample step signal as the frequency domain features, divides the spectrum of a sample step signal into several frequency bands, as well as calculates each frequency band’s amplitude standard deviation and constructs a vector, serving as a training(testing) sample. Simultaneously, fuzzy SVM method is introduced; fault membership degree of fault detection samples is calculated by constructing fault membership functions. Furthermore, the failure membership is used to eliminate misclassification fault detection samples, so as to realize control of the algorithm to control risks of false alarms, and then the accuracy of algorithms is raised.Tests proved the above two turbopump fault detection algorithms both meet requirements for validating algorithms’ accuracy, timeliness and real-time, they are of certain effects and positive meaning for improving test safety of the liquid rocket engine’s turbopump and reducing the fault loss of turbopump.