| Summary: | The digital transformation of production systems enables the production of large volumes of multivariate time series, which in turn facilitates the continuous monitoring of production assets. The modelling of multivariate time series can reveal the way parameters evolve as well as the influence amongst themselves. These data can be used in tandem with artificial intelligence methods to create insight upon the condition of production equipment, hence, potentially increasing the sustainability of existing production sites, by reducing waste and production downtime. In this context, LSTM-Autoencoders and a Transformer encoder are investigated in this study to enable predictive analytics through spatial and temporal time series forecasting. These neural networks are applied in combination upon a real-world dataset, targeting an efficient model for RUL estimation. The aforementioned methodology is implemented into a software prototype which is trained and tested in a real-world steel industry case.
|
|---|
