Research On Key Techniques Of Health Management In Liquid Propellant Rocket Engine Ground Testing Bed

On the various stages of the rocket engine development, we need to do a lot of ground tests to solve complex problems. Rocket engine ground test-bed system has the large scale and complex structure. Due to the numberous measurement parameters, accurate measurement class, expensive test costs and dangerous experiment, test failure will result in huge economic losses. In order to improve the reliability and security of the test-bed, and reduce launch costs, national scholars began to research on test-bed health management technology. Based on a certain liquid rocket engine ground test-bed, this paper researched on fault detection, fault diagnosis and health evaluation of health management and assessment. According to the different characteristics of failure modes in the test-bed, the paper studied the corresponding real-time detection and fault diagnosis algorithm, and established the health assessment system to slove many core technical issues of test-bed health management.To resolve the problem of liquid rocket engine test-bed fault detection, this paper proposed a fault-online monitoring method, which based on wavelet kernel KPCA algorithm, to overcome the traditional approach ignoring the correlation parameters in solving nonlinear process defects. Through wavelet kernel achieving high-dimensional mapping, originally nonlinear problem was convered into a linear problem in high-dimensional space. This is a novel access to address the problem of test-bed fault detection. And by using KPCA detection methods it achieved test-bed fault detection. This method is very suitable for non-linear deterministic fault detection.To solve the problem of liquid rocket engine test-bed fault diagnosis, this paper proposed a Rough set-Relevance Vector Machine(RVM) fault diagnosis. In this section, the paper studied the rough set based attribute reduction method. Though test-bed original knowledge systems feature extraction, this method can improve efficiency of test-bed fault diagnosis and greatly reduct the system decision table. Additionally, the paper focuses on multi-level relevance vector machine classification based on clustering algorithm. The key of this method is the use of adaptive particle swarm optimization(APSO) to optimise each two sorter model parameters of multi-sorter and the use of standard data to verify the performance of the niche particle swarm optimization. Simulation results show that rough set attribute reduction method can effectively extract failure characteristics types. This method is superior to the traditional classification methods in the efficiency calculation and fault identification degree. After model optimization, RVM multiple classifier has better overall performance and it can effectively solve the small sample problem of system fault diagnosis.According to requirements of the liquid rocket engine test-bed health assessment, a novel strategy by using relevance vector machine(RVM) coupled with fuzzy comprehensive evaluation method is proposed for the health evaluation and prognosis of liquid engine test-bed system. Expanding the application on the engines test-bed system, include health evaluation of single sensor parameter, multiple fuel and oxid izer subsystems, and the whole test-bed system. The HRD concept is reemployed to evaluate the health evaluation of test-bed system in a quantitative way. The novelty of this paper lies in the improvement of the HRD methodology itself by using the RVM-based multi-variable data fusion technology coupled with fuzzy evaluation algorithm. This strategy has combined the advantages of RVM-based good generalization performance under small sample and fuzzy theory-based comprehensive evaluation, in which the health prognosis and HRD fusion are accomplished.Finally, this paper designed and implemented the test-bed health management system, and simulated the live environment on PC. By using simulation data and test-bed real data this paper verified the correctness and effectiveness of the fault detection, fault diagnosis and health assessment algorithm.