Deep Learning Models Applied to Intelligent Diagnosis of Rotating Machines

AI algorithms can help detect anomalies and identify failure modes under specific conditions, making them valuable tools in maintenance management. However, there is no consensus on which of them is the most effective because each author builds a different architecture based on the main deep learning models, changing functions, parameters and normalizations, and with different databases, making a fair comparison between the models impossible. To address this issue, this work proposes a brief review using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) methodology to identify works in the literature that perform intelligent diagnostics on available datasets of rotating machines using deep learning algorithms. After this review, this work also presents new results from the use of models, such as multilayer perception (MLP), auto-encoder (AE), convolutional neural network (CNN) and recurrent neural network (RNN), making direct comparisons of the result obtained with the outcomes found in the literature after the review. To support the discussions about the results, confusion matrix, accuracy and losses' graphs were generated for all combinations between models and input types applied.

Keywords: Intelligent diagnosis, Reliability, Prognostic health management, Deep learning, Rotating machines.

Download PDF