Predictive maintenance PdM has helped, in recent decades, manufacturing and industry to save costs and keep their operations safe. This study outlines how advanced machine learning systems, including LSTMs and Transformers, could enhance data-driven maintenance planning. Mainly using C-MAPSS datasets to test Deep Learning (DL) methods and estimate Remaining Useful Life (RUL) values, this research aims to compare LSTM networks and Transformer performance in prognostic via different evaluation criteria. The analysis manifests the peculiarities of both the proposed learning approaches with a marked difference in the performances in favor of the recurrent architecture (e.g., 40% in term \(R^2\) and 43% in terms of MSE), thus not generally suggesting the usage of transformer-based architecture, especially in a data-scarcity situation common condition when working of critical and costly units, but opening to a new perspective in otherwise operational conditions where data is prosperous.