Available at: http://digitalcommons.calpoly.edu/theses/1515
Date of Award
MS in Aerospace Engineering
Inductive Monitoring Systems (IMS) are the newest form of health monitoring available to the aerospace industry. IMS is a program that builds a knowledge base of nominal state vectors from a nominal data set using data mining techniques. The nominal knowledge base is then used to monitor new data vectors for off-nominal conditions within the system. IMS is designed to replace the current health monitoring process, referred to as model-based reasoning, by automating the process of classifying healthy states and anomaly detection. An IMS prototype was designed and implemented in MATLAB. A verification analysis then determined if the IMS program could connect to a CubeSat in a testing environment and could successfully monitor all sensors on board the CubeSat before in-flight use. This program consisted of two main algorithms, one for learning and one for monitoring. The learning algorithm creates the nominal knowledge bases and was developed using three data mining algorithms: the gap statistic method to find the optimal number of clusters, the K-means++ algorithm to initialize the centroids, and the K-means algorithm to partition the data vectors into the appropriate clusters. The monitoring algorithm employed the nearest neighbor searching algorithm to find the closest cluster and compared the new data vector with the closest cluster. The clusters found were used to establish the knowledge bases. Any data vector within the boundaries of the clusters was deemed nominal and any data vector outside the boundaries was deemed off-nominal. The learning and monitoring algorithms were then adapted to handle the data format used on a CubeSat and to monitor the data in real time. The developed algorithms were then integrated into a MATLAB GUI for ease of use. The learning and monitoring algorithms were verified with a 2-dimensional data set to ensure that they performed as expected. The final IMS CubeSat prototype was verified using 56-dimensional emulated data packages. Both verification methods confirmed that the IMS ground- based prototype was able to successfully identify all off-nominal conditions induced into the system.