Machine Learning and Neural Network for Maintenance Management

The paper presents a novel approach that allows to optimize the ultrasonic wave sensors for a condition monitoring system employing. It can detect and diagnosis different faults with a signal, such as delamination, mud or ice on blades of wind turbines. This methodology allows to avoid the redundancy of sensors, since a specific number of ultrasonic transducers can determine the structural condition using guided waves. The signal is pre-processed with the aim of removing the noise, then extracted and selected features to be later classified by Machine Learning and Neural Networks. Finally, for each damage or anomaly, the best classifier will be evaluated. The best classifier of each damage will act on a parallel network that will process the signal sent by the sensor.