Vibration Analysis of 3D printed runner with CNN for using deep learning in hydropower for condition monitoring

Abstract

Convolutional Neural Networks are deep learning algorithms that are typically used for the classification of images. A black and white image can be represented as the number of rows and columns along with values of intensity for each of components of these rows and columns. Vibration measured with the help of accelerometer can be used to create an image like data. These data can be used for machine learning applications with the help of deep learning. In deep learning techniques, features do not have to be extracted from the data set. The CNN network can generate a feature-like set which can be further classified with a fully connected neural network to give out the probability over the predefined class. The neural networks model is trained with a data set created in laboratory conditions to monitor the unfractured and single blade fracture condition operation state. The paper deals with showing the proof of concept that CNN algorithms can be applied for conditional monitoring of runner blades. The paper mainly focuses on demonstrating that deep learning algorithms can be used as analysis tools for hydropower runners by applying these techniques in scaled versions of 3D printed runners.