Automating Rotational Machinery Diagnostics using Multimodal Deep Learning and Developing Cloud-based Online Model Training Systems

Abstract

With full cooperation from experienced diagnosticians and engineers, we have developed a system that automates rotational machinery diagnostics using multimodal deep learning method. The multimodal approach has advantage in mimicking diagnosticians’ thinking process and works well with current deep learning frameworks. We discuss how the multimodal model is built and how it behaves on robustness testing. We also introduce originally developed wireless acceleration sensor device and cloud-based application which allows us to update model online.