Towards Autonomous Predictive Maintenance: A Bibliometric Review of Machine Learning Approaches and a Self-Learning Agentic AI Framework

In today’s competitive marketplace, minimising downtime, expensive repairs, and increasing operational efficiency have made predictive maintenance (PdM) a critical component of an effective industrial system strategy. Industry 4.0, along with the rising availability of sensor data, has spearheaded the utilisation of machine learning (ML) technology in developing sophisticated solutions for predictive maintenance. By examining numerous scientific papers using the inclusion and exclusion criteria, this work investigates what is known and not known in the domain of the research, as well as key challenges, research trends, and future developments. The analysis also pinpoints the leading authors, institutions, and sources of information, as well as the history of applying ML techniques (such as supervised, unsupervised, and deep learning) for predictive maintenance. In addition, the paper addresses challenges and opportunities as well as possibilities of implementing machine learning based predictive maintenance systems in real conditions, such as data representativeness, transparency of machine learning models, and deployment in a real-time environment. This work meticulously presents a systematic bibliometric analysis of the intersection between predictive maintenance and industrial machines using machine learning algorithms. This review proposed a novel self-learning agent-based predictive maintenance framework capable of autonomy to make decisions with minimum human assistance involving orchestrations of maintenance AI agents, thus taking a step towards a fully adaptive and intelligent industrial system.