5G-Enabled NetApp for Predictive Maintenance in Critical Infrastructures Sofia Giannakidou K3Y Ltd, Sofia, Bulgaria sgiannakidou@k3y.bg Panagiotis Radoglou-Grammatikis University of Western Macedonia, K3Y Ltd Kozani, Greece - Sofia, Bulgaria pradoglou@uowm.gr, pradoglou@k3y.bg Sotirios Koussouris Suite5 Data Intelligence Solutions Ltd, Limassol, Cyprus, sotiris@suite5.eu Minas Pertselakis Suite5 Data Intelligence Solutions Ltd, Limassol, Cyprus, minas@suite5.eu Nikolaos Kanakaris Public Power Corporation S.A. Athens, Greece nickanakaris@gmail.com Alexios Lekidis Public Power Corporation S.A. Athens, Greece A.Lekidis@dei.gr Konstantinos Kaltakis Eight Bells Ltd, Nicosia, Cyprus, konstantinos.kaltakis@ 8bellsresearch.com Maria P. Koidou University of Macedonia, Thessaloniki, Greece maria koidou@yahoo.gr Chrysi Metallidou University of Macedonia, Thessaloniki, Greece chrysi.metallidou@gmail.com Konstantinos E. Psannis University of Macedonia, Thessaloniki, Kastoria, Greece kpsannis@uom.edu.gr Sotirios Goudos Aristotle University of Thessaloniki, , Greece sgoudo@physics.auth.gr Panagiotis Sarigiannidis University of Western Macedonia, Kozani, Greece psarigiannidis@uowm.gr Abstract—Predictive Maintenance in critical infrastructure is a fundamental tool for predicting a failure in advance and for avoiding catastrophic equipment damage that can be prevented and the time-consuming repair scheduling can be executed in time. Artificial Intelligence (AI) based predictive maintenance utilises intelligent data for accurate predictions in order to make immediate interventions on critical assets. In this paper, we propose a 5G-enabled Network Application (NetApp) for predictive maintenance in energy-related critical infrastructures. The proposed NetApp consists of several containerised com- ponents responsible for retrieving time-series operational data from a power plant and detecting potential outliers/anomalies regarding the operation of energy generators. For the anomaly detection process, an autoencoder is used. The evaluation results demonstrate the efficiency of the proposed NetApp. Index Terms—5G, Artificial Intelligence, Network Application Autoencoder, Predictive Maintenance I. I NTRODUCTION Predictive Maintenance is a crucial maintenance tool based on the possibility of estimating the future values of some quantities that characterise a system (typically a machine, a plant, or a production process) through particular mathematical models in order to identify in advance the anomalies and po- tential faults. The predictive maintenance applications predict failure sufficiently ahead of time in order for the decision- makers to take appropriate actions, such as maintenance, replacement or even a planned shutdown. These applications promote savings on machine maintenance and increase produc- tivity by guaranteeing the maximum uptime of machines. The manufacturing processes mostly adhere to an assembly line production, therefore any failure in the assembly line results in a domino effect, making it essential to evade any point of failure within the assembly line. By deploying predictive maintenance solutions, these failures can be evaded or at least minimised. However, for the most accurate and optimal prediction, it is required to gather and analyse large amounts of relevant data within a reasonable time frame. Consequently, big data analytics and stream processing technologies are key necessities for predictive maintenance solutions. Predictive maintenance applications are acknowledged as one of the fundamental data-driven analytical applications for large-scale manufacturing industries. In this paper, we provide a Network Application (NetApp) for predictive maintenance in power plants, taking full advan- tage of containerisation, 5G and Artificial Intelligence (AI). In particular, the proposed NetApp adopts an autoencoder in order to recognise timely potential anomalies/outliers with respect to the functionality of industrial devices, paying special attention to electricity generators. For this purpose, operational data of the electricity generators are used. This kind of data is received through Programmable Logic Controllers (PLCs). Next, the autoencoder receives this kind of data and is respon- sible for the detection of anomalies. The rest of this paper is organised as follows. Section II provides some relevant works in this field. Next, section III describes the architecture of the proposed NetApp. Next, section IV focuses on the evaluation of the autoencoder. Finally, section V concludes this report. II. RELATED WORK Several works investigate predictive maintenance applica- tions and models in critical infrastructures, such as [1]–[5]. XXX-X-XXXX-XXXX-X/XX/$XX.00 ©20XX IEEE