AI-Enabled Digital Twin Framework for Predictive Maintenance and Performance Optimisation of Mechatronic Mechanical Systems

Abstract

The growing integration of artificial intelligence and digital twin technologies into predictive maintenance has greatly contributed to the optimisation of the performance of mechatronics and industrial systems; yet several challenges, related to model validation, scalability, and real-world deployment, still pose serious concerns. The current paper describes a systematic literature review (SLR) carried out based on the PRISMA approach with the purpose of analysing the latest progress made in AI-based modelling techniques used for predictive purposes. Peer-reviewed academic publications on the use of AI-based techniques published from 2018 to 2025 were chosen from Scopus, Web of Science, ScienceDirect, and SpringerLink online databases. The findings show that there is a prevalent use of machine learning and deep learning approaches such as Artificial Neural Networks (ANN), Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), Support Vector Machines (SVM) and physics-informed models in combination. These algorithms are used mainly for detecting faults, predicting RUL, analysing anomalies, and conducting predictive analytics. The high level of predictive accuracy achieved, nevertheless, is accompanied by insufficient research in real-world implementation, IIoT applications, and system validation. The existing models are mostly experimental and require improvements in terms of robustness, generalizability, data quality, and validation process. The paper gives an overview of modelling techniques used, approaches to validation, and application areas, highlighting the most relevant problems faced during real-world deployment and the lack of integration with digital twins.