However, these approaches can be costly and time-consuming. Predictive maintenance (PredM) offers a more proactive and cost-effective solution by leveraging data analytics to predict equipment failure before they occur. In this short article, we will present the concept of predictive maintenance as it applies to power systems and in particular in the context of I-NERGY (an ERC H2020 funded project).

PredM essentially involves collection and analysis of data from various sensors installed in the equipment/system (Munteanu et al., 2022) under consideration to identify deviations from normal operating conditions. The collected data is then used to identify potential issues in the equipment/system before they ultimately result to a breakdown or outage. Adopting PredM can help power systems operators to significantly reduce equipment downtime, improve reliability, and extend the life of their assets.

There are several techniques that can be employed in PredM; they include vibration analysis, oil analysis, thermography etc. (López-Pérez and Antonino-Daviu, 2017). Vibration analysis involves measuring the vibrations of rotating equipment to detect any abnormal patterns that could result in failure of the equipment. Oil analysis involves examining the properties of lubricating oils in the asset to detect any changes that might be indicative of equipment wear or contamination. Thermography involves measuring the temperature of equipment to determine any hot spots that may indicate excessive friction or electrical resistance than the expected value obtained during normal operation of the equipment.

PredM like other technical fields can be improved by use of machine learning (ML) algorithms. These algorithms can be trained to identify abnormal patterns in the data and predict equipment failure in advance. ML can also help to optimize maintenance schedules by identifying the most critical equipment as well as predicting optimal time to perform maintenance.

Good PredM practice in power systems can result in huge benefits. For instance, By identifying potential issues before they become critical, power system operators can significantly reduce equipment downtime and thus improve overall reliability of the system. PM can also help to reduce maintenance costs by enabling more efficient and targeted maintenance interventions.

It is worthy of note that lack of adequate data can seriously hamper PredM service development. This is the case with the I-NERGY project, where lifetime data of the assets from the project pilots are not available. Thus, RWTH Aachen University (one of the I-NERGY project partners) is developing PredM service for EPES assets using degradation approach. The approach consists of the ML part as well as PredM service development part. The following figure depicts these steps. The ML part consists of two-step process: (i) prediction of when the EPES asset or system is going to fail and (ii) identification of the asset(s) that will contribute more towards an imminent failure of the system. The first aspect is handled using an ensemble of ML models while the second part is handled using a variational auto-encoder (a generative model). The second part (service development) involves deployment of the models in addition to scripts used to generate PredM schedule which is accessible via a dashboard by the operators. The first part of this task is mostly accomplished, the current (ongoing task) is the finalization of the service development and testing at the pilot sites.

From the foregoing, we summarize that PM is a valuable tool for power system operators to improve the efficiency and reliability of their systems. By leveraging data analytics and ML, operators can effectively identify potential equipment issues before they cause breakdowns or outages. Further, this proactive approach to maintenance can bring about significant cost savings and improved operational performance of EPESs.