Research Of Sensor Fault Diagnosis And Data Reconstruction For Testbed

The liquid rocket engine ground test is very complex. The testing system adopted in the test has the features of large scale, complex architecture, large quantity of sampling data. During the test, so much volatile and burnable propellant is used that the process is dangerous and influenced by many factors. As an important part of testing system, the sensor, which needs high reliability to assure the security of ground test, inspects the system state in real time. So the studies on diagnostic methods for sensor fault are valuable very much.This paper focuses on two diagnostic methods for sensor fault: principal component analysis (PCA) and artificial neural network predictor. Three kinds of pressure sensors from hydrogen system are employed to lucubrate PCA .The experiments illuminate that PCA is effective for typical sensor fault. It is proved that artificial neural network predictor studying on-line possesses capabilities of robustness and adaptability, which ensure its high precision and real-time performance.The reconstruction method to fault data is studied in this paper. The reconstruction with PCA gets excellent outcome. The Elman artificial neural network predictor is adopted to reconstruct fault data. Because the space redundancy information between sensors is fused, better reconstruction result is achieved. With the performance of higher precision and speed, the two methods can be used in real time for data reconstruction.At last, the sensor fault diagnostic software of liquid rocket engine ground test system is developed successfully, including the designs of arithmetic, database and interface software. Finally, the applicability and effectiveness of the method studied are illustrated by the software designed.