Artificial neural network based modeling of SSME sensor signals

摘要

Recent space shuttle main engine (SSME) condition monitoring activities have focused on both real-time safety monitoring and post-test diagnostics. In order to prevent shutdowns or other controller actions due to sensor failures, a real-time sensor validation system is critical. Techniques are being developed to automate the post-test diagnostic procedure. The long term goal of this research is to develop an artificial neural network based fault detection system which successfully identifies and differentiates between engine and sensor failures. The objective of the current research is to develop and evaluate artificial neural network (ANN) based signal approximation models for use in SSME sensor validation. In order to develop SSME sensor signal approximation model, a set of sensor signals must be selected for use as input to each ANN model. Based on the set of input parameters, the neural network provides an estimate of the sensor signal being modeled. The sensor signal approximation models could be used for sensor fault detection and isolation during ground test firings. Further, such a signal approximation model would provide an estimate of a critical sensor signal which could be used for continued monitoring and analysis in the event of a sensor failure. If accurate and reliable models are available, future engines can continue to operate using a synthesized signal. This is especially attractive for space engines where the entire engine is anticipated to be a LRU (Line Replaceable Unit). The predicted signals can also be used to enhance post-test diagnostic evaluations and life prediction calculations.