Detailed description of the project. 

The phenomena associated with Lightning Events (LE) represent a critical issue for the Electrical Infrastructures (EI). Among them, LE mostly impact the Transmission and Distribution Systems (TS & DS), but also Renewable Energy Sources (RES) and electronic devices are sensible to such damages. When dealing with LE, it is possible to distinguish between two categories: direct hits and indirect strokes (the lightning hits the ground in the proximity of the EI). While direct hits are much more dangerous, indirect ones have a higher occurrence probability, which leads to a greater number of faults and damages. Typical examples of sensitive systems are overhead power lines of both TS and DS, Photovoltaic (PV) plants and Wind Turbine (WT) systems, all commonly positioned in large open spaces, hence more exposed to the lightning risk.

Present lightning protection standards provide criteria for designing different kinds of Lightning Protection System (LPS), aimed at mitigating the risk associated with LE for the components of the EI. LPSs are based on the risk assessment, i.e., on the combination of:

• the number of lightning events affecting the considered EI (threat); 
• the probability that a lightning event that affects the considered EI causes damage (vulnerability);
• the amount of the associated loss (consequence).

The protection of EI from LE [1] is based on the idea that hits are unavoidable due to the random nature of LE, and it is only possible to reduce damages. According to IEC EN 62305, four lightning protection levels can be introduced and the possible protection measures depend on the infrastructure to be protected, but can be summarized as follows:

• Shielding wires (for DS and TS) [2]
• Earthing and Bonding measures [3]
• Surge protective devices [4]
• Cables shielding [5]

The choice of the protection system involves three main subsequent phases: a) the lightning current modeling [6], b) the Electro-Magnetic Fields (EMF) modeling [7]–[8] and c) the modeling of the EMF coupling with the infrastructure to be protected, which provides the effective damage inferred to it.

The evaluation of EMF coupling has been faced in several papers, analyzing the possible damages inferred to TS and DS [4], [9], [10] as well as on the RES (especially PV plants and WT) [11]–[14], leading to different models to evaluate the effectiveness of mitigation/protection systems. Due to the complexity of the EI, the most adopted models are the numerical ones, typically divided into transmission-line models and full-wave models. While the methods belonging to the first category usually require low computational effort, the second ones are capable of modeling the complete electromagnetic coupling mechanism in soil or air with a higher level of detail.

A LE is associated to different phases [15]: first of all, the lightning initiates as a corona discharge caused by the overcoming of the air insulating capacity, which leads to the formation of the so called “streamers”, cold corona filaments of the length of few meters. The second phase is the transition of the streamers to the formation of the so-called “stepped leader”, which is a self-propagating discharge in an electric field that produces a hot plasma channel with a continuing current. The stepped leader, while propagating downward, carries down a negative/positive charge. Finally, when the stepped leader gets near the ground (about 100 meters), other streamers move from the ground up towards the stepped leader and, when the stepped leader and one of the upward streamers attach, a high pulse of energy flows up towards the cloud and towards the ground. This luminous pulse is the Return Stroke (RS) and represents the most critical part of the lightning in terms of damages to electric and electronic devices. Up to now, the design of the protection systems have always been based on the analysis of the RS.

Different works [16]–[18] studied the modeling of the initial phases of LE (i.e., the so called Preliminary Breakdown Pulses - PBP) in terms of measured electromagnetic fields and in terms of induced voltages on DS, showing that the leader electric field has a significant impact on the amplitude and shape of the induced voltages in the time instants preceding the return stroke [18]. The authors of [16] have investigated the possible relationship between the electric field pulses and the preliminary phases of lightning discharge, while in [18] it is shown that in some cases the leader-induced voltages can be large enough to trigger protection devices.
 

It has been observed that PBP have generally much less energy and evolve on a much slower dynamics than RS. While the damage occurrence associated with the preliminary phases of LE is not completely known and can be neglected in many cases, it is important to note that those signals have never been used as preliminary detectors of the return stroke. The current state of the art of lightning protection measures does not include any preventive or predictive protection able to predict the occurrence of a dangerous RS starting from the PBP detection and to disconnect the electronic equipment of the RES.
 

The project FELINES aims at filling this gap, investigating the possibility of finding a predictive signal of the RS, providing a new type of protection measure. For direct lightning strikes, LPS usually provide an effective protection, since the lightning current path is more predictable. Indirect lightning strike effects are much more difficult to contrast, as different (and often unpredictable) paths transfer the energy from the lightning strike to the electronics inside the systems. There are basically two mechanisms that allow energy from a strike to couple with a device indirectly. These are:

• Near-field and far-field electromagnetic coupling; lightning events produce electromagnetic waves, which couple with metallicparts. There are two simple factors which gauge how much energy is transferred: i) the proximity to the lightning strike and ii) the interested area of the conductor (e.g., the length of a cable run).
• Ground potential gradient coupling; this coupling mechanism works through the interaction of the lightning strike and the earth (e.g., dirt, soil, etc.) around impact points. The impact produces a current from the strike that travels on earth outwards from that point. These currents produce voltage-potential perturbations on the earth's surface, and systems having multiple connections to earth can be susceptible to energy coupling through those grounding points from ground-potential gradients induced by lightning
  currents. Therefore, lightning protection for EIs needs to be improved.
   

The key idea of the FELINES project could reduce the number of direct events that cause dangerous overvoltages and provide an effective protection also against damages from indirect hits. PBP have been studied in different works [25]–[28], where measured PBP data have been processed to analyze their relationship with the RS and models of the PBP have been proposed to reproduce the measured phenomenon in terms of current, electromagnetic fields, and induced voltages on a DS.
 

Although the authors of [28] showed that in some cases the PBP induced voltages can be large enough to trigger protection devices, the risk associated with the preliminary phases of lightning discharge is typically not considered. This can be reasonable, since their direct inclusion in the risk assessment procedures would result in a significant increase of complexity, without an analogous increase of benefits. Nevertheless, PBP are no doubt a resource that should be exploited in lightning research. In [25]-[27], details of sensors able to detect the EMFs associated with PBP are made available. 'Early' fields could be detected using, e.g., broadband antennas combined with Ultra High Frequency (UHF) antenna arrays [27]. Moreover, with respect to the RS, the initial parts of the lightning process are associated to much slower dynamics (in the order of tens or hundreds of microseconds), and
much smaller currents in the lightning channel (in the order of few kiloamperes), at least until just before the RS. Moreover the typical time interval between PBP and RS of some tens of milliseconds [25]-[27]. Therefore, early PBP detection allows a protection triggering signal for the disconnection of all vulnerable and precious electronics
(i.e., power converters) immediately before the lightning-induced overvoltage develops could be designed. This is the final goal of FELINES project. Being aware that limited resources and duration allow just to investigate the feasibility of the concept, the main goal is to provide a “Proof Of Concept” (POC) as the final outcome of the research.

Moreover, this research project aims at releasing specifics about the required instrumentation. In particular: what kind of sensors are needed, in what range of frequency they should operate, what are the proper RC (resistor/capacitor) time constants for all different sensors, what are the proper sampling time and record length, what are the tolerable detection errors, what is the correct sensors number and placement for the different electrical infrastructures that will be considered.

1. The detection of phenomena related to PBP from field sensors located close to the apparatus to be protected is not trivial for two different reasons: the signal emitted by PBP is weaker with respect to that of the RS and can be affected by noise
2. the characteristics of the electromagnetic fields associated with the PBP have not been studied as deeply as the ones related to the RS.
    

In order to cope with the noisy and unpredictable nature of typical waveforms associated with PBP, we plan to adopt Machine Learning (ML) algorithms and tools to recognize specific patterns that can represent reliable prediction of future damage. More in details, the detection of markers related to PBP in measured EMF can be treated as a typical classification problem, i.e., a specific type of pattern recognition. A classification problem attempts to assign each input value to one of a given set of classes. In the system to be designed, the input values are the PBP signals (electromagnetic field waveforms), whereas the classes are “the incoming RS will be/will not be dangerous for the EI to be protected”.
 

Unfortunately, the experimental data are rather scarce, so it is not possible to establish a completely experimental data set for the training and testing processes required in any ML method. Moreover, to the present day, no data on PBP current and electromagnetic field waveforms are available for Italy. However, it has long been demonstrated that computational analysis of electromagnetic fields can be applied to model the interaction of electromagnetic fields generated by LE with objects and the environment using Maxwell’s equations. Therefore, the FELINES project aims at developing a model that, starting from a LE, is able to reproduce the PBP EMFs, the RS EMFs and their coupling with the object to be protected. Using numerical simulations, a large number of “virtual” experiments will be generated, and an optimized detection system will be designed. The signals from the probes
will then be used to train and test a ML system to detect incoming RS as soon as possible.
 

The three Research Units (RU) involved in FELINES have complementary skills but, at the same time, can have also some experience in the research field relevant to the other Units. The RU of UNIGE has a long and well recognized experience in the modelling and study of the lightning phenomenon and how it affects electrical infrastructures, with particular attention dedicated to WT and Ts&DS overhead lines. For this specific expertise the researchers of this RU are able of understanding the outcomes of each phase of the project, consequently the RU of UNIGE will
coordinate the whole project.
 

The RU of UNIPI has been very active in the last years in the development of preventive and predictive maintenance procedures based on advanced data mining techniques (applied to PV systems and electric traction systems), and in the development of ML based optimization procedures for electromechanical devices. For this reason this RU key role will be the development of the early detection algorithm.
 

The Research Unit of UNICAMP has a long experience in the advanced modelling of complex electromagnetic systems; lately the researchers of this RU have dedicated their attention to the modeling and evaluation of lightning effects on PV plants and in the use of ML based algorithm for the optimization of electromagnetic devices and the solution of inverse problems.
 

• More in details, during recent years, the RUs proposing this project developed useful tools dealing with the RS analysis which can be exploited and extended to the PBP in order to reach the goals of the project. In particular, the research group has the availability of: a full-wave model of lightning RS affecting a WT to study the lightning current reflections and distortion and the variations in the electromagnetic fields due to the presence of the WT [20] (UNIGE)
• efficient methodologies for the evaluation of lightning-induced overvoltages on overhead transmission and distribution lines (e.g., [29] and [30]) (UNIGE and UNICAMP)
• a model of PV module and DC/DC converter assembly for the analysis of induced transient response due to nearby lightning strike [16] and a model for the assessment of induced voltages in common and differential-mode for a PV module due to nearby lightning strike [17] (UNICAMP)
• numerical models suited to estimate the coupling parameters between lightning electromagnetic fields and metallic structures [31] (UNICAMP and UNIPI)
• A solid experience on the use of ML to cope with direct and inverse problems in Electromagnetism (UNIPI)
   

As it is evident from the above list and from the works published by the coordinators of each RU, this group is characterized by strong and complementary technical skills. This allows the group to be able to carry out the project objectives in a cost and time efficient way; at the same time the partial overlapping between the RU competences guarantees an efficient cooperation and common ground, for an optimal and smooth running of the project.
 

In order to cope with the foreseen amount of work, the RU of UNIGE plans to hire a researcher with a research grant (12 months), and will also exploit additional man-power from young researchers already belonging to the RU. The RU of Pisa will hire a researcher with a 18 months research grant. 

Therefore, the challenge to address consists in performing a complete overview of the models for the estimation of the PBP current that are available in the literature, analyzing and comparing them, and assessing their robustness and limits of validity. Preliminary tests will be carried out to test the performance of different PBP models. Once one or more PBP models are selected, they will be implemented in order to simulate the measurements from the selected sensors/antenna. As a further development, PBP models will also be integrated into the available tools previously developed by the RUs for the assessment of the RS effects on WT systems [20], overhead TS&DS power lines (e.g., [29] and [30]), and PV plants [16] [17] to foresee possible damages from PBP events.
 

Moreover, the relationship between PBP and RS features will be investigated. Some analyses are available in the literature comparing PBP and RS parameters. The goal is to understand how to estimate the RS current, starting from the measured PBP electromagnetic field.
 

Successively, a significant number of simulations will be performed. The goal of this crucial phase of the project is to generate an appropriate data set that, starting from a simulated PBP electromagnetic signal, produces as final output the answer to the question “will the device to be protected be damaged by the incoming RS?”. Of course, inside the black box between the input and the output there are all the conceptual steps which have been discussed in the previous paragraphs, and summarized in the following: i) regression from simulated electromagnetic PBP signals to the PBP current, through the selected model, ii) estimation of the RS current from the PBP current, iii) evaluation of the electromagnetic fields generated by the estimated RS, and iv) coupling of such electromagnetic fields with the object to be protected in order to discriminate whether damages occur or not. It is stressed that points iii and iv of the previous list will be addressed through the available tools developed by the RUs (see [16], [17], [20], [29], [30] and [31]) and properly adapted.
 

Preliminary considerations of the characteristics of the data set will be performed to optimize the process. In particular, the following issues will be objects of investigation: the proper size of the data set, the range of variability of the involved parameters, the special distribution of simulations (i.e., to decide at what distances from the object to be protected the lightning events will be simulated), the sampling time and so on.
 

From the data set generated in the previous step, an advanced data mining will be performed. The goal of this phase is to set-up a proper ML based algorithm which can work as the desired classifier. To address this point, the applicability of different ML based algorithms for signal processing, clustering and pattern recognition will be investigated. Preliminary tests will be done to select the most appropriate technique, which, probably, will result from a combination of different ML based algorithms. In fact, the final aim of the FELINES project is the design of a robust system, able to properly identify and prevent any possible source of damage with negligible classification errors.
 

To complete this final step of FELINES, the competencies developed by the RU of UNIPI in the field of Artificial Intelligence (AI) and ML [32]-[33] will be exploited to perform a rigorous and efficient set-up of the algorithm. Moreover, note that the computational efficiency of the LE detection tool represents a fundamental constraint. Indeed, the classifier will be able to process the incoming PBP signal, develop an appropriate output, and eventually send a device disconnection command in a time window much shorter than the typical time interval between the first preliminary pulse and the RS, which usually consists of tens of milliseconds [26], [27]. The integration of the tools to be developed in the project with the already available one and the role of each RU in the project framework is depicted below.

FELINES project does involve three main advancements of knowledge. First of all, due to the limited number of research studies on the PBP with respect to the works on the RS phase, and on the coupling of related EMFs with EI, FELINES aims at improving the knowledge on the PBP. Moreover, the new knowledge in terms of PBP will be transferred to the lightning analysis tools implemented in different commercial softwares (EMTP, PSCAD-EMTDC, Simulink), which are usually based only on the RS phase.

As a final, yet not less relevant, advancement, the project will reverse the idea that stands behind the LPS presently installed. While standard LPS are based on limiting the damage inferred by LE with the installation of surge arresters, shield wires and groundings, FELINES introduces the concept of preventive protection, i.e. the preliminary disconnection of the electronic equipment when a probable dangerous event is detected.

BIBLIOGRAPHY

[1] V. Cooray, Lightning protection. 2009.
[2] M. Brignone, D. Mestriner, R. Procopio, F. Rachidi, and A. Piantini, ‘Mitigation of Lightning-Induced Overvoltages Using Shield Wires: Application of the Response Surface Method’, in 2018 34th International Conference on Lightning Protection (ICLP), 2018, pp. 1–6.
[3] M. Paolone, C. A. Nucci, E. Petrache, and F. Rachidi, ‘Mitigation of lightning-induced overvoltages in medium voltage distribution lines by means of periodical grounding of shielding wires and of surge arresters: Modeling and experimental validation’, IEEE Trans. Power Deliv., vol. 19, no. 1, pp. 423–431, 2004.
[4] D. Mestriner, R. A. Ribeiro de Moura, R. Procopio, and M. A. de Oliveira Schroeder, ‘Impact of Grounding Modeling on Lightning-Induced Voltages Evaluation in Distribution Lines’, Appl. Sci., vol. 11, no. 7, p. 2931, 2021.
[5] M. A. O. Schroeder, M. T. C. de Barros, A. C. Lima, M. M. Afonso, and R. A. Moura, ‘Evaluation of the impact of different frequency dependent soil models on lightning overvoltages’, Electr. Power Syst. Res., vol. 159, pp. 40–49, 2018.
[6] A. Borghetti, C. A. Nucci, M. Paolone, M. Bernardi, S. Malgarotti, and I. Mastandrea, ‘Influence of surge arresters on the statistical evaluation of lightning performance of distribution lines’, in 2004 International Conference on Probabilistic Methods Applied to Power Systems, 2004, pp. 776–781.
[7] M. Nazari, R. Moini, S. Fortin, and F. P. Dawalibi, ‘On the Performance of Grounding, Bonding, and Shielding Systems for Direct and Indirect Lightning Strikes’, IEEE Trans. Electromagn. Compat., vol. 62, no. 1, pp. 135–143, Feb. 2020
[8] F. Heidler, J. M. Cvetic, and B. V. Stanic, ‘Calculation of lightning current parameters’, IEEE Trans. Power Deliv., vol. 14, no. 2, pp. 399–404, 1999.
[9] M. Brignone, D. Mestriner, R. Procopio, and F. Delfino, ‘A review on the return stroke engineering models attenuation function: Proposed expressions, validation and identification methods’, Electr. Power Syst. Res., vol. 172, pp. 230–241, 2019.
[10] M. Brignone, R. Procopio, D. Mestriner, M. Rossi, F. Delfino, F. Rachidi and M. Rubinstein, ‘Analytical Expressions for Lightning Electromagnetic Fields With Arbitrary Channel-Base Current—Part I: Theory’, IEEE Trans. Electromagn. Compat., vol. 63, no. 2, pp. 525-533, April 2021
[11] A. Shoory, A. Mimouni, F. Rachidi, V. Cooray, R. Moini, and S. H. Sadeghi, ‘Validity of Simplified Approaches for the Evaluation of Lightning Electromagnetic Fields Above a Horizontally Stratified Ground’, IEEE Trans. Electromagn. Compat., vol. 52, no. 3, pp. 657–663, Aug. 2010, doi: 10.1109/TEMC.2010.2045229.
[12] D. Li, M. Azadifar, F. Rachidi, M. Rubinstein, M. Paolone, D. Pavanello, S. Metz, Q. Zhang and Z. Wang, ‘On lightning electromagnetic field propagation along an irregular terrain’, IEEE Trans. Electromagn. Compat., vol. 58, no. 1, pp. 161–171, 2015.
[13] Y. Baba and V. A. Rakov, ‘Electromagnetic models of the lightning return stroke’, J. Geophys. Res. Atmospheres, vol. 112, no. D4, 2007.
[14] ‘IEEE Guide for Improving the Lightning Performance of Electric Power Overhead Distribution Lines’, IEEE Std 1410-2004 Revis. IEEE Std 1410-1997, pp. 1–50, Jul. 2004.
[15] M. Brignone, F. Delfino, R. Procopio, M. Rossi, and F. Rachidi, ‘Evaluation of power system lightning performance - Part II: Application to an overhead distribution network’, IEEE Trans. Electromagn. Compat., vol. 59, no. 1, pp. 146–153, 2016.
[16] A. Formisano, J. C. Hernández, C. Petrarca, and F. Sanchez-Sutil, ‘Modeling of PV Module and DC/DC Converter Assembly for the Analysis of Induced Transient Response Due to Nearby Lightning Strike’, Electronics, vol. 10, no. 2, Art. no. 2, Jan. 2021.
[17] A. Formisano, C. Petrarca, J. C. Hernández, and F. J. Muñoz-Rodríguez, ‘Assessment of induced voltages in common and differential-mode for a PV module due to nearby lightning strikes’, IET Renew. Power Gener., vol. 13, no. 8, pp. 1369–1378, 2019.
[18] J. C. Hernandez, P. G. Vidal, and F. Jurado, ‘Lightning and surge protection in photovoltaic installations’, IEEE Trans. Power Deliv., vol. 23, no. 4, pp. 1961–1971, 2008.
[19] N. I. Ahmad, M.Z.A. Ab-Kadir, M. Izadi, M.A.M. Radzi, N.H. Zaini and M.S.M. Nasir, ‘Lightning protection on photovoltaic systems: A review on current and recommended practices’, Renew. Sustain. Energy Rev., vol. 82, pp. 1611–1619, 2018.
[20] R. Torchio, M. Nicora, D. Mestriner, M. Brignone, R. Procopio, P. Alotto, M. Rubinstein et al., ‘Do Wind Turbines Amplify the Effects of Lightning Strikes A Full-Maxwell Modelling Approach’, IEEE Trans. Power Deliv., 2022.
[21] F. Rachidi, M. Rubinstein, J. Montanyà, J. Bermudez, R. Sola, G. Sola and N. Korovkin, ‘A review of current issues in lightning protection of new-generation wind-turbine blades’, IEEE Trans. Ind. Electron., vol. 55, no. 6, pp. 2489–2496, 2008.
[22] IEC 61400-24:2019. Wind Energy Generation Systems – Part 24: Lightning Protection
[23] G. V. Cooray, The lightning flash. Iet, 2003.
[24] V. Cooray, An introduction to lightning. Springer, 2015.
[25] P. Kašpar, O. Santolík, I. Kolmašová, and T. Farges, ‘A model of preliminary breakdown pulse peak currents and their relation to the observed electric field pulses’, Geophys. Res. Lett., vol. 44, no. 1, pp. 596–603, 2017.
[26] I. Kolmasova, T. Marshall, S. Bandara, S. Karunarathne, M. Stolzburg, N. Karunarathne and R. Seidlecki, ‘Initial breakdown pulses accompanied by VHF pulses during negative cloud-to-ground lightning flashes’, Geophys. Res. Lett., vol. 46, no. 10, pp. 5592–5600, 2019.
[27] Y. Wang, X. Qie, D. Wang, M. Liu, D. Su, Z. Wang, D. Liu, Z. Wu, Z. Sun and Y. Tian, ‘Beijing Lightning Network (BLNET) and the observation on preliminary breakdown processes’, Atmospheric Res., vol. 171, pp. 121–132, 2016.
[28] F. Rachidi, M. Rubinstein, S. Guerrieri, and C. A. Nucci, ‘Voltages induced on overhead lines by dart leaders and subsequent return strokes in natural and rocket-triggered lightning’, IEEE Trans. Electromagn. Compat., vol. 39, no. 2, pp. 160–166, 1997.
[29] M. Brignone, F. Delfino, R. Procopio, M. Rossi and F. Rachidi, ‘Evaluation of Power System Lightning Performance, Part I: Model and Numerical Solution Using the PSCAD-EMTDC Platform’, in IEEE Transactions on Electromagnetic Compatibility, vol. 59, no. 1, pp. 137-145, Feb. 2017
[30] D. Mestriner, M. Brignone, R. Procopio, M. Nicora, E. Fiori, A. Piantini and F. Rachidi, ‘An Efficient Methodology for the Evaluation of the Lightning Performance of Overhead Lines’, IEEE Trans. Electromagn. Compat, vol. 63, no. 4, pp. 1137-1145, Aug. 2021
[31] A. Formisano, J C. Hernandez, C. Petrarca and F. Sanchez-Sutil, ‘Computation of Coupling Parameters between Neighboring Panels’ in Photovoltaic Arrays, Proc. Of Compumag2021, Cancun (Mexico), 16-20 Jan. 2022.
[32] S. Barmada, M. Raugi, M. Tucci and F. Romano, ‘Arc detection in pantograph-catenary systems by the use of support vector machines-based classification’, IET Electrical Systems in Transportation, vol. 4, no. 2, pp. 45 - 52, 2014
[33] A. Betti, M. Tucci, E. Crisostomi, A. Piazzi, S. Barmada and D. Thomopulos, ‘Fault Prediction and Early-Detection in Large PV Power Plants Based on Self-Organizing Maps’, Sensors, vol. 21, no. 5 , pp. 1–16, 2021.
[34] H. K. Hoidalen, ‘Calculation of lightning-induced overvoltages in MODELS including lossy...