Fault Detection and Diagnosis in Propulsion Systems: A Fault Parameter Estimation Approach

Introduction T HERE is a growing demand to improve the control systems of liquid propulsion rocket engines for enhanced performance with increased reliability, durability, and maintainability. This demand can be met by improving the individual reliabilities of system components and also by an intelligent control system with fault detection, diagnostics, and accommodation capabilities. This paper focuses on the development of a model-based fault detection and diagnosis (FDD) system that can be used as an integral part of such an intelligent control system. During the last two decades of the development of fault detection methods, the so-called model-based fault detection approach has received considerable attention. These schemes basically rely on the idea of analytical redundancy. As opposed to physical redundancy, which uses measurements from redundant sensors for fault detection purposes, analytical redundancy is based on the signals generated by the mathematical model of the system being considered. These signals are then compared with the actual measurements obtained from the system. The comparison is done by using the residual quantities that give the difference between the signals being measured and the signals being generated by the mathematical model. Hence, the model-based fault detection and diagnosis can be defined as the determination of faults of a system from the comparison of the measurements of the system with a priori information represented by the model of the system through generation of residual quantities and their analysis. In the absence of noise and modeling errors, the residual vector is equal to the zero vector under fault-free conditions. Hence, a nonzero value of the residual vector indicates the existence of the faults. When noise and modeling errors are present, their effect has to be separated from the effect of faults. In the simplest case, this is done by comparing the residual magnitudes with threshold values. Using the distribution of the residuals under fault-free conditions, one can determine threshold values to minimize false alarms and missed detections by selecting the level of confidence.