A neural network for the analysis of test data from the aeropropulsion systems test facility

With the advent of the USAF's Advanced Tactical Fighter and NASA's National Aerospace Plane, demands for concise test data reduction and interpretation will increase beyond the capabilities of current methodologies. As mission complexity increases it becomes apparent that real time data analysis for flight safety, mission control and test conduct becomes a necessary tool. A neural network is a biologically inspired mathematical model, which can be represented by a directed graph, that has the ability to learn through training. Neural networks have many advantages over current aviation computing systems including the ability to learn and generalize from their environment. Neural networks are excellent for parameter estimation and recognizing patterns in signal data. This thesis discusses a prototype system designed and implemented at the University of Tennessee Space Institute to discover patterns in test data from an engine test cell in order to determine if any part of the system is in failure. The results of this research show that a neural network can be used for fault diagnosis in an engine test cell when the problem of fault monitoring and diagnosis is seen as one of pattern recognition. A two layer semilinear feed-forward neural net is able to separate simulated sensor data into normal and abnormal classes and the addition of a hidden layer makes the network more resistant to noise and improves the ability of the network to classify the type of fault that is occuring. Neural networks offer test personnel a new tool for the analysis of critical test data. A general working network model can be applied to any system where sensor data is evaluated and fault diagnosis is critical. Significant benefit to the aerospace community is gained by improved safety of test engineers by rapid detection of faulty conditions. This will be of extreme importance as testing begins on more advanced engine systems such as the National Aerospace Plane and the Advanced Tactical Fighter.