The distribution network is recognized for its inherent fragility and challenging management compared to the transmission network. This challenge arises from the complex topology of the distribution network, involving thousands of nodes to monitor by predictive maintenance. This paper introduces the implementation of an artificial neural network (ANN) for predictive maintenance of medium voltage (MV) switchgears. In particular, the study here presented consists in proposing a new approach that correlates temperature measurements, at different positions within the switchgear, to potential faults. The specific cause of temperature change is the variable fastening in a single point of the MV switchgear. Thus, starting from experimental measurements, the ANN is thoroughly analyzed, compared, and validated to provide a proper classification of the switchgear health status. The obtained results affirm the efficacy of the proposed approach and highlight the benefits of its application in practical predictive maintenance scenarios.
Electrical utilities and system operators (SOs) are constantly looking for solutions to problems in the management and control of the power network. For this purpose, SOs are exploring new research fields, which might bring contributions to the power system environment. A clear example is the field of computer science, within which artificial intelligence (AI) has been developed and is being applied to many fields. In power systems, AI could support the fault prediction of cable joints. Despite the availability of many legacy methods described in the literature, fault prediction is still critical, and it needs new solutions. For this purpose, in this paper, the authors made a further step in the evaluation of machine learning methods (ML) for cable joint health assessment. Six ML algorithms have been compared and assessed on a consolidated test scenario. It simulates a distributed measurement system which collects measurements from medium-voltage (MV) cable joints. Typical metrics have been applied to compare the performance of the algorithms. The analysis is then completed considering the actual in-field conditions and the SOs’ requirements. The results demonstrate: (i) the pros and cons of each algorithm; (ii) the best-performing algorithm; (iii) the possible benefits from the implementation of ML algorithms.
This paper presents a distributed measurement system intended to effectively monitor the health status of switchgears under varying temperature conditions. In particular, thermocouples are deployed as temperature sensors for the continuous monitoring of a medium-voltage (MV) switchgear. Then, by integrating a low-cost microcontroller unit, the proposed system can implement previously trained unsupervised learning techniques for health status evaluation. This approach enables the early detection of potential faults by identifying anomalous temperature patterns, thus supporting predictive maintenance and extending the lifespan of switchgears. The results show strong clustering performance with low execution times, highlighting the suitability of the method for resource-constrained hardware. Furthermore, onboard temperature processing eliminates the need for data transmission to remote servers, reducing latency and communication overhead while improving system responsiveness. The paper includes a numerical analysis on synthetic data as well as a validation on real measurements. Overall, the presented distributed measurement system offers a scalable and cost-effective solution to enhance the reliability and safety of MV switchgears.
12 March 2025 in IEEE Transactions on Instrumentation and Measurement
The distribution network relies on a critical electric component, the cable joint (CJ), for expansion and post-fault repairs. In some parts of the network, a CJ can be found every few hundred meters, making it a crucial element that requires thorough study. Additionally, the CJ is highly vulnerable, and its failure often results in explosions following a fault in a problematic network branch. Unfortunately, the exact relationship between CJ parameters and their health index (HI) is unknown. This work aims to enhance existing knowledge by proposing a new and flexible HI and using a machine learning-based approach to estimate CJ aging. The proposed HI considers numerous factors, detailed in the text, and is adaptable to the information available to system operators. By means of a public dataset with condition monitoring data, the machine learning algorithm is validated to demonstrate its efficacy in estimating the HI and forecasting future CJ parameters.
Recent environmental concerns have heightened attention toward new solutions across all fields to mitigate human impact. The power system community is also deeply committed to addressing this issue, with research increasingly focused on sustainable practices. For instance, there is a growing trend in designing new buildings to be net-zero emitters, while older structures are being retrofitted for energy efficiency to achieve similar goals. To this purpose, the study aims to enhance the energy management capabilities of an educational building by implementing a smart infrastructure. Equipped with photovoltaic panels and a distributed measurement system, the building captures voltage and current data and calculates power. These electrical quantities are then forecasted through an AI-driven framework that manages the data. The paper details the AI model used, including its experimental validation. The results show that the system provides reliable forecasts of electrical parameters. The evaluation of the distributed measurement system and the collected data offers valuable insights, which support more informed actions for optimizing energy management and system performance. A key novelty of this study lies in the exploration of model generalization across measurement nodes. This approach is supported by the correlation analysis of data, which highlights the potential for accurate predictions in case of data gaps. Moreover, the ease of deployment and the practical application of the system were highlighted as key factors for scalability, allowing for potential adaptation in similar infrastructures.
23 February 2026 in IEEE Transactions on Instrumentation and Measurement
Diabetic retinopathy (DR) is a major cause of vision impairment worldwide, and early diagnosis remains a key challenge. Building upon previous work that demonstrated the utility of machine learning (ML) in classifying DR patients based on electromyographic (EMG) muscle response to visual stimuli, this study extends the investigation by analyzing inter-trial dynamics, test duration efficiency, and subject-aware cross-trial transferability. A Random Forest (RF) classifier was trained on muscle response data collected from healthy individuals, untreated DR patients, and laser-treated patients, each exposed to structured visual stimuli across multiple trials. The study evaluates classification performance under varying conditions: different acquisition lengths, reduced test durations, inter-trial transfer scenarios within the same subject, and in the presence of measurement uncertainty. Results confirm that patient class and stimulus type can be reliably predicted even with short-duration acquisitions, highlighting that early EMG signal segments encode sufficient discriminative information. Additionally, performance trends across trials provide insights into response stability and intra-subject adaptation effects over time. This work reinforces the potential of ML-assisted neuromuscular analysis in DR diagnosis within controlled, subject-aware measurement scenarios and suggests pathways for more efficient, personalized screening protocols.