Alex Rubenstein

Self-Organizing Map Neural Networks and Rule Extraction Techniques

Date: 2/27/2019
Time: 4:30PM-5:30PM
Place: 315 Armstrong Hall

Abstract: An artificial neural network (ANN) has been implemented as a key component in the development of an on-board health-state awareness technology to predict degradation of an unmanned aerial vehicle (UAV). The standard electronic speed control on a small UAV has been replaced with an Intelligent Electronic Speed Control (IESC) that uses telemetry data from sensors to develop an intelligent rule set extracted from a trained ANN to monitor system degradation and predict specific types of faults. The dynamic cell structure (DCS) ANN is changed through Hebbian and Kohonen learning to allow for adaptation to new inputs with the goal of learning the topology of the input space. Extraction of the knowledge the ANN has learned in a more understandable format provides fixed rule sets that may be implemented in a system to perform in a similar manner to the ANN after training.

All are welcome.