A Temporal-causal-graph-based Fault Diagnosis Method For Liquid-propellant Rocket Engines

Paper number

IAC-07-C4.I.10

Author

Dr. Jianjun Wu, National University of Defense Technology, China

Coauthor

Year

2007

Abstract

Like most of current qualitative fault diagnosis methods, ambiguity always exists in the results of causal- -graph-based diagnosis method because three kinds of important information are always neglected. The first is the dynamic information of the system such as the trend in the change of symptoms. The second is the propagation duration between the nodes in causal graph models. The third is the importance of the occurred faults because the most possible faults should be considered foremost. Thus the causal-graph –models-based diagnosis method lacks the ability of real-time detection and diagnosis in the dynamic working process such as the startup process which is always critical for liquid-propellant rocket engines (LRE). 
So a new diagnosis method for LRE based on the Temporal-Causal-Graph (TCG) is developed in this paper. 
Firstly, the definition of the TCG, the relation and difference between the TCG based diagnosis method and other general causal graph based diagnosis method are all analyzed and clarified. Different from general causal graphs, an attribute vector is added in TCG to represent the dynamic behavior of engines such as the derivatives of variables, the propagation duration of faults, and so on. 
Secondly, a new modeling method is developed to build the TCG model of engines from the little-deviation dynamic equations of LRE. By this way, the complexity of modeling the dynamic characteristic of LRE and the difficulty of building the dynamic TCG model from bond graphs are both avoided, as most causal graph based methods did. 
Thirdly, the framework of fault diagnosis based on the TCG model for LRE is proposed which includes the assumption generation of faults, the prediction of behavior and the fault diagnosis. A backward propagation algorithm is first used to generate assumptions of faults. Then the dynamic behaviors of the engine are predicted based on the propagation network in TCG in a way of forward propagation. At last, a progressive monitoring strategy is used to check the consistency and produce the final diagnosis results. By this way, the generated unreasonable assumptions of faults will be eliminated and the ambiguity of the results will be reduced gradually with the increase of measured information.
The method is verified by simulated fault samples of a liquid-propellant rocket engine example. Results show that the method proposed is effective. It can be available for reference in the design and realization of practical fault diagnosis systems.

Abstract document

IAC-07-C4.I.10.pdf

Manuscript document

IAC-07-C4.I.10.pdf (🔒 authorized access only).

To get the manuscript, please contact IAF Secretariat.