"Deep Neural Networks for Gear Degradation Prediction in Mechanical Systems: An Analysis of Multi-Data Streams"

This thesis work aims to explore and develop a convolutional neural network to address a predictive maintenance problem. The objective is to accurately estimate the wear and tear status of a gear component within a complex system, created through a controlled laboratory experiment. The dataset comprises nine data streams obtained from three accelerometers at a high sampling frequency. The problem is approached as both a classification and regression task, with various CNN models being developed to address it. The study's major result is the development of ensemble learning techniques, which deliver noteworthy accuracy. This thesis work contributes to the field of predictive maintenance, offering insights into the capabilities and limitations of CNN models in solving complex problems in industrial settings.