June 2020

AI Deep Learning Methodology: The Convolutional Neural Network applied for Remaining Useful Life Image Classification

Abstract: The advent of Industry 4.0 introduced new methods for maintenance that enable faster and reliable use of data for classification. The so-called “Convolutional Neural network” can be applied for classification and regression. Concerning classification, the use of images can be applied to train and test a model to classify similar images such as equipment failure during production, wrong assembly, equipment failure or degradation during operation or even detect critical degradation pattern based on Remaining Useful Life (RUL) and/or State of Health (SoH) graphs. In order to apply the CNN methodology, is necessary to collect a set of pictures that aims to classify in the future application based on CNN structures. The “Convolutional Neural Network (CNN)” is like neural network in principle but differs in concept. Since the CNN aims to classify images, the structure of the CNN is quite complex with different types of layer as shows the figure below. The elements of CNN are the following:

  • Input Image
  • Convolution Layer
  • Pooling Layer
  • ReLu
  • Fully Connected
  • Softmax
  • Output Classification