Automated parameter optimization for feature extraction for condition monitoring

Abstract

Pattern recognition and signal analysis can be used to support and simplify the monitoring of complex aircraft systems. For this purpose, information must be extracted from the gathered data in a proper way. The parameters of the signal analysis need to be chosen specifically for the monitored system to get the best pattern recognition accuracy. An optimization process to find a good parameter set for the signal analysis has been developed by the means of global heuristic search and optimization. The computed parameters deliver slightly (one to three percent) better results than the ones found by hand. In addition it is shown that not a full set of data samples is needed. It is also concluded that genetic optimization shows the best performance.