AI for Prognostics: a Survey on Emerging Approaches and Implementation

The increasing demand for reliability, safety, and efficiency across complex engineering systems has placed Prognostics and Health Management (PHM) at the centre of industrial innovation. In particular, accurately estimating Remaining Useful Life (RUL) is crucial for predicting failures and extending system lifecycles. The progressive integration of Artificial Intelligence (AI) into PHM has reshaped the way reliability and maintenance are addressed in engineering systems emerging as a key enabler of advanced prognostic methods and complementing or surpassing traditional physics-based models. This thesis presents a systematic literature review (SLR) on emerging approaches for AI-based prognostics, analysing fifty-four peer-reviewed studies across multiple engineering domains including aerospace, mechanical equipment, manufacturing, wind energy, and automotive. The review identifies clear trajectories ranging from physics-based and model-driven approaches, based on deterministic equations, to data-driven approaches based on machine learning and deep learning architectures, and finally toward hybrid solutions in which physical knowledge and AI models are integrated. Deep Learning techniques such as Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), Long Short-Term Memory (LSTM) networks, autoencoders, Echo State Networks (ESN), and more recently Transformers, are compared in terms of their methodological principles, applications, strengths, and limitations. Particular attention is devoted to macrotrends such as the widespread adoption of deep architectures and the growing emphasis on uncertainty quantification, explainability, and trustability. By mapping the state of the art and reflecting on their limitations, this paper contributes to a deeper understanding of how AI can be effectively integrated into PHM and sketches the direction in which future research and industrial adoption are expected to move.